T Todd ElvinstoddacmorgIndicast Corporation

Carlsbad CA 92008

David R NadeaunadeausdsceduSan Diego Supercomputer Center

Rina SchulrschulcrlucsdeduSDSUUCSD Joint Doctoral

Program in Clinical Psychology

David KirshkirshcogsciucsdeduCognitive Science DepartmentUniversity of California San Diego

Presence Vol 10 No 6 December 2001 565ndash582

copy 2001 by the Massachusetts Institute of Technology

Worldlets 3-D Thumbnailsfor Way nding in LargeVirtual Worlds

Abstract

Finding onersquos way to sites of interest on the Web can be problematic and this dif-culty has been recently exacerbated by widespread development of 3-D Web con-

tent and virtual-world browser technology using the Virtual Reality Modeling Lan-guage (VRML) Whereas travelers can often navigate 2-D Web sites based ontextual and 2-D thumbnail image representations of the sitesrsquo content nding onersquos

way to destinations in 3-D environments is notoriously troublesome Wayndingliterature provides clear support for the importance of landmarks in building a cog-nitive map and then using that map to navigate in a 3-D environment be it real or

virtual Textual and 2-D image landmark representations however lack the depthand context needed for travelers to reliably recognize 3-D landmarks This paperdescribes a novel 3-D thumbnail landmark affordance called a worldlet Containing a

3-D fragment of a virtual world worldlets offer travelers rst-person multi-view-point experience with faithful representations of potential destinations

To facilitate an investigation into the comparative advantages of landmark affor-dances for waynding worldlet capture algorithms were designed implementedand incorporated into two VRML-based virtual environment browsers Findings

from a psychological experiment using one of these browsers revealed that com-pared to textual and image guidebook usage worldlet guidebook usage nearly dou-bled the time subjects spent studying the landmarks in the guidebook signicantly

reduced the time required for subjects to reach landmarks and reduced backtrack-ing to almost zero These results support the hypothesis that worldlets facilitatetraveler landmark knowledge expedite waynding in large virtual environments and

enable skilled waynding

1 Introduction

As the Web moves into 3-D the problem of nding onersquos way to sites ofinterest is exacerbated URLs currently are represented on a page or menu bysentences as in gure 1 or by 2-D thumbnail images that depict some aspectof the URL as in gure 2 It is reasonable to suppose that interactive 3-Dthumbnailsmdash or worldletsmdashwill soon come into use To test whether worldletsimprove waynding we implemented a prototype 3-D system and designed apsychological experiment aimed at answering two questions

c Are menus of worldlets better than menus of text and 2-D images for way-nding

Elvins et al 565

c What aspects of the waynding process are mostenhanced by the use of worldlets

In the experiment subjects were placed at startinglocations in three virtual cities and asked to nd theirway to goal locations via a series of landmarks In onecity the landmarks were described with text in anothercity the landmarks were described using 2-D thumbnailimages and in yet another city the landmarks were de-scribed using manipulable worldlets as shown in gure3 We hypothesized that compared to text and thumb-nail images worldlets would facilitate better landmarkknowledge and expedite waynding

2 Way nding

Waynding is ldquoa term that can refer to a rather nar-row concern That is how well people are able to ndtheir way to their particular destination without delay orundue anxietyrdquo (Peponis Zimring amp Choi 1990 p561) Whereas navigating is most often dened as theprocess of following a course by making directed move-ments waynding is the critical process of determining

the strategy direction and course needed to reach adesired destination Travelers nd their way using sur-vey procedural and landmark knowledge (Goldin ampThorndyke 1982 Peponis et al 1990) Each type ofknowledge helps the traveler construct a cognitive mapof a region and thereafter nd their way using that map(Appleyard 1976 Downs amp Stea 1973 Lynch 1960Passini 1992)

Survey knowledge provides a map-like birdrsquos-eyeview of a region and contains spatial information includ-ing locations orientations and sizes of regional fea-tures Procedural knowledge characterizes a region bymemorized sequences of actions that construct routes todesired destinations Landmark knowledge records thevisual features of landmarks including their 3-D shapesize texture and so forth (Appleyard 1976 Goldin amp

Figure 1 A Web page linking to alternative destinations within a

virtual world On this page the destinations are represented with text

Figure 2 On this Web page the destinations from Figure 1 are

represented as 2-D thumbnail images

566 PRESENCE VOLUME 10 NUMBER 6

Thorndyke 1982) For a structure to be a landmark itmust have high imagability that is it must be distinctiveand memorable (Lynch 1960)

Landmarks are the subject of landmark knowledgebut they also play a part in survey and proceduralknowledge In survey knowledge landmarks provideregional anchors with which to calibrate distancesand directions In procedural knowledge landmarksmark decision points along a route helping in therecall of procedures to get to and from destinationsof interest

Whereas a travelerrsquos landmark knowledge charac-terizes a destination by its 3-D shape size textureand so forth traditional virtual environment (VE)browser menus and landmark guidebooks character-ize destinations only by textual descriptions or im-ages This representation mismatch reduces the effec-tiveness of destination menus and guidebooksUnable to engage their memory of 3-D landmarksto recognize destinations of interest travelers mayresort to a naive exhaustive search to nd a desiredlandmark

Figure 3 Construction of a worldlet A gas station landmark in a virtual city (top left) viewed from a

viewpoint in the street (top right) Viewed from above the gas station is upper left and a region

around the previous viewpoint is highlighted (bottom left) The gas station is captured within a spherical

worldlet and (bottom right) is viewed from street level within the worldlet

Elvins et al 567

21 Related Work

Although social and engineering scientists havestudied human waynding in the real world for manyyears (Appleyard 1976 Downs amp Stea 1973 Goldinamp Thorndyke 1982) only recently have computer sci-entists investigated this subject Satalich (1995) led sub-jects on a tour of a virtual building and then tested thesubjectsrsquo ability to nd their way to places seen duringthe tour Darken and Sibert (1996) applied Lynchrsquos(1960) city structure theories to large sparsely featuredVEs and then compared the subjectsrsquo ability to ndtheir way to target destinations with and without theadded city structure Witmer Bailey Knerr and Parsons(1996) studied route learning and found that subjectswho rehearsed a route in a virtual building learned theroute better than did subjects who rehearsed the routeverbally Gillner and Mallot (1997) found that subjectspieced together survey knowledge when repeatedly re-quired to nd shortest paths between a dozen unfamil-iar locations in a VE In studying VEs for real-worldtask training Waller Hunt and Knapp (1998) foundthat long exposure fostered good spatial knowledge un-der several VE delity conditions Ruddle Payne andJones (1998) studied path complexity and a compasscondition as factors in VE route learning and foundthat beyond two turns subjects were quickly disori-ented with or without a compass Although each ofthese studies recognized the value of landmark knowl-edge for waynding none studied the value of 3-Dlandmarks in familiarizing subjects with an environmentbefore entering it This is a primary goal of our study

22 Way nding Affordances

Three-dimensional models of large cities are be-coming available for exploration via the Web(Art 1 Com 1997 Liggett Friedman and Jepson1995 University College London 1997) As discussedpreviously travelers visiting such large VEs for the rsttime are easily disoriented may have difculty identify-ing a place upon arrival and may not be able to ndtheir way back to a place just visited (Downs amp Stea1973 Darken amp Sibert 1996) The following subsec-

tions describe affordances that travelers currently use toscrutinize potential landmark and VE destinations

c Textual descriptions are the dominant method usedto represent VE landmarks in world- and viewpoint-selection user interfaces HTML pages for instanceoften provide lists of available Web-based worlds(such as those authored in the Virtual Reality Mod-eling Language (VRML) (ISO 1997)) each onecharacterized by a URL a VE name andor a briefdescription Within VRML worlds textual descrip-tions characterize viewpoints and describe destina-tions associated with clickable anchor shapes

c Icons thumbnail images and image maps providecommon visual waynding aids In a 3-D contextgames often provide ldquojump gatesrdquo onto which im-ages of remote destinations are texture mappedStepping through such a gate jumps the traveler tothe destination landmark depicted on the gate

c Image mosaics group together multiple capturedimages into a traversable structure Applersquos Quick-Time VR for instance can use images capturedfrom multiple viewing angles at the same viewingposition (Chen 1995) By ordering images within atraveler-centered cylindrical structure QuickTimeVR can provide a traveler the ability to look in anydirection through automatic selection of an appro-priate image from the structure By chaining multi-ple mosaic structures together the content authorcan create a walkthrough path that hops from view-point to viewpoint Similar image mosaics can beused to create zoom paths pan paths and so forth

c Miniature Worlds and Maps are available in most3-D environment browsers and computer gamesand they enable the traveler to zoom out and viewthe world in miniature and to thereby gain surveyknowledge Pausch Burnette Brockway andWeiblen (1995) extend this notion by creating aworld in miniature (WIM) embedded within themain world The miniature world duplicates all ele-ments of the main world and adds an icon denotingthe travelerrsquos position and orientation Held withinthe travelerrsquos virtual hand the traveler can reachinto the miniature and reposition world content

568 PRESENCE VOLUME 10 NUMBER 6

andor themselves Simultaneously the outer mainworld is updated to match the altered miniatureautomatically adjusting the positions of shapes orthe traveler Similarly 2-D and 3-D maps are fre-quently found as navigation aids within VEs Forinstance 3-D games often provide a 3-D reduced-detail map in which an icon denotes the playerrsquoslocation Such maps can be panned zoomed androtated to provide alternate viewpoints similar tothat which is possible with miniature worlds

None of these waynding affordances offer the trav-eler faithful 3-D landmark representations or a rst-person experience with potential destinations Wickensand Baker (1995) argue that VE search tasks are dif-cult because ldquooften the object of the search is speciedin a format different from and more abstract than itsrendering in the VErdquo We address this problem by rep-resenting the object of the search in the same 3-D for-mat as the VE

3 What are Worldlets

In previous papers we described our ongoingworldlet waynding research (Elvins 1997 ElvinsNadeau amp Kirsh 1997 Elvins 1998 Elvins NadeauKirsh amp Schul 1998) Similar to a 3-D photograph aworldlet is associated with a viewing position and orien-tation and captures a snapshot of the 3-D shapes (usu-ally landmarks) that fall within the viewpointrsquos viewingvolume Worldlets can be explored and manipulated inthe same way that virtual worlds can be explored andmanipulated Incorporated into a VRML browser aguidebook or list of worldlets enables a world traveler tointeractively view landmarks and landmark context frommultiple viewpoints Seeing the landmarks presented inworldlets helps a traveler select destinations of interestand nd their way in unfamiliar worlds

In typical use the worldletrsquos viewpoint is aimed at animportant landmark and the worldletrsquos captured shapesreconstruct that landmark and its associated contextWhen viewed within an interactive 3-D browser a

worldlet provides a manipulable 3-D thumbnail repre-sentation of the landmark

Clicking on a worldlet in a landmark menu mightjump the traveler to the landmark represented by theworldlet The landmark could be at a viewpoint in thecurrent world or in another world A menu of worldletssuch as that shown in gure 4 might resemble land-mark history lists bookmarks or travel guidebooksTravelers are free to study the landmark in the worldletfrom any viewpoint before deciding whether to gothere

We have developed two types of worldlets

c A frustum worldlet contains shapes within a stan-dard pie-shaped viewing frustum positioned andoriented based upon a selected viewpoint Whenviewed a frustum worldlet looks like a pie-shapedfragment clipped from the larger main world

c A spherical worldlet contains shapes within a spheri-cal viewing bubble positioned at a selected view-point with a 360 deg eld of view When dis-played a spherical worldlet looks like a ball-shapedworld fragment similar to a snowglobe knick knackdepicting a favorite attraction

Figure 4 The navigation browserrsquos viewpoint selection window and

the waynding browserrsquos worldlet guidebook The ldquoGo tordquo button was

removed in the waynding browser

Elvins et al 569

For both worldlet types hither and yon clipping planesrestrict the extent of the worldlet ensuring that theworldlet contains a manageable subset of the largermain world Worldlet shape content is shaded and tex-tured to match the corresponding shapes in the mainworld Although the main world may have content thatchanges over time the captured worldlet remains staticrecording the content of the world at the time theworldlet was captured

Figure 3 depicts the process of constructing a spheri-cal worldlet from a fragment of a virtual city surround-ing a central viewpoint position Once captured world-lets can be browsed manipulated and explored in anumber of ways

4 Worldlets in Browsers

We have incorporated worldlets into the user in-terface for two experimental VRML browsers a naviga-tion browser and a waynding browser The navigationbrowser provides a ldquoGo tordquo feature that enables travel-ers to study select and y to world viewpoints in thecurrent world and to study select and load previouslyvisited worlds on the browserrsquos history list Guidebooksof worldlets in the waynding browser allow travelers tostudy 3-D landmark guidebooks before and duringtravel in an unfamiliar VE No go-to button is presentin this browser in which travelers manually move them-selves about based on the landmark knowledge gleanedduring study

41 Navigation Browser

Traditional VRML browsers provide a viewpointmenu that offers a choice of viewpoints each denotedby a brief textual description We have extended thisstandard feature to provide three experimental view-point selection interfaces each using worldlets All threepresent a set of worldlets one for each author-selectedviewpoint in the world A ldquoGo tordquo button ies the mainwindowrsquos viewpoint to that of the currently selectedworldlet The browser also supports on-the-y captureof worldlets from the travelerrsquos current viewpoint

c The viewpoint list window provides a list of world-lets beside a worldlet viewer Selection of a worldletfrom the list displays the worldlet in the viewerwhere it may be interactively panned zoomed androtated

c The viewpoint guidebook window presents a grid ofworldlet viewers arranged to form a guidebookphoto-album page Buttons on the window advancethe guidebook forward or back a page at a timeEach worldlet on the page can be interactively ex-amined Figure 4 shows the viewpoint guidebookwindow

c The viewpoint overlay window enables the travelerto select a worldlet from a list and overlay it atopthe main window highlighted in green This world-let overlay provides a clear indication of the world-letrsquos viewpoint position and orientation along withthe portion of the world captured within thatworldlet Figure 3b was generated using this overlaytechnique

Traditional VRML browsers provide a history list ofrecently visited worlds each denoted by its title orURL We have extended this standard feature to pro-vide two world selection interfaces each using world-lets In each world selection window a ldquoGo tordquo buttonloads into the main window the world associated withthe currently selected worldlet

c The world list window provides a list of worldletsbeside an interactive worldlet viewer similar to theviewpoint list window discussed earlier One world-let is available for each world on the browserrsquos his-tory list

c The world guidebook window uses the same guide-book photo-album layout used for the viewpointguidebook window discussed earlier One worldletis available for each world on the history list

A ldquoSave asrdquo feature of the VRML browser enables thetraveler to save a worldlet to a VRML le Using a col-lection of saved worldlets a world author can create aVRML world of worldlets Such a world acts like a 3-Ddestination index similar to a shelf full of snowglobeknick knacks When cast as a VRML anchor shape a

570 PRESENCE VOLUME 10 NUMBER 6

worldlet provides a 3-D button that when clicked onloads the associated world into the travelerrsquos browser

Figure 5 shows a world of worldlets Figure 5a shows

a close-up view of a world ldquodoorwayrdquo and a niche thatcontains a worldlet illustrating a viewpoint in thatworld Figure 5b shows a wider view of the same worldand multiple such doorways

42 Way nding Browser

The second browser was implemented specicallyfor the psychological experiment After manipulatingworldlets in a landmark guidebook to familiarize them-selves with the 3-D size shape color texture and con-text of the landmarks subjects in the experiment foundtheir way manually in the main window based upontheir landmark knowledge The landmark guidebookresembled the viewpoint selection window shown ingure 4 but it did not offer a ldquoGo tordquo button Thewaynding browser is described in detail in the methodssection

5 Implementation

Worldlets can be implemented in several waysThe implementation presented here is based on ahardware-accelerated algorithm and support algorithmsdeveloped specically for fast execution The VRMLbrowser used in this work maintains VE geometrywithin a tree-like scene graph Worldlets are also storedas scene graphs together with additional state informa-tion To capture a worldlet or to display a worldlet orVE the VRML browser traverses the associated scenegraph and feeds a 3-D graphics pipeline

51 Worldlet Capture in General

Any 3-D graphics pipeline can be roughly dividedinto two stages (i) transform clip and cull and (ii)rasterize (Foley van Dam Feiner amp Hughes 1990)The rst stage applies modeling viewing perspectiveand viewport transforms to map 3-D shapes to the 2-Dviewport Along the way shapes outside of the viewingfrustum are clipped away and backfaces are removedThe second stage uses 2-D shapes output by the rst

Figure 5 A world of worldlets that (top) associates a worldlet with

each doorway (bottom) in an environment containing multiple such

doorways Each doorway leads to a different world

Elvins et al 571

stage and draws the associated points lines and poly-gons on the screen

Worldlet capture taps into this 3-D graphics pipelineextracting the transformed shaded clipped and culledshape coordinates output by the rst stage prior to ras-terization in the second stage An extracted coordinatecontains x and y screen-space components a depth-buffer z-space component and the w coordinate Eachextracted coordinate has an associated RGB color andtexture coordinates which are computed by shadingand texture calculation phases in the rst pipeline stage

To create a worldlet these extracted coordinates areuntransformed to map them from viewport space backto world space as shown in gure 6 The inverses of the

viewport perspective viewing and modeling transformsare each applied Coordinate RGB colors and texturecoordinates are used to reconstruct 3-D worldlet geom-etry in a worldlet scene graph

Display of a worldlet passes this 3-D geometry backdown the graphics pipeline transforming clipping cull-ing and rasterizing the worldlet like any other 3-D con-tent

52 Frustum and Spherical Worldlets

A frustum worldlet is the result of capturing 3-Dgraphics pipeline output for a single traversal of thescene graph as viewed from the travelerrsquos current view-point The shape set extracted after the rst pipelinestage contains only those points lines and polygonsthat fall within the viewing frustum The worldlet con-structed by the browser from this geometry looks like apie-shaped slice cut out of the world

A spherical worldlet is the result of performing multi-ple frustum captures and combining the results TheVRML browser captures a spherical worldlet by spin-ning about a viewpoint position generating a set offrustum worldlets that each use a different viewing ori-entation Additional captures aimed straight up andstraight down complete the spherical worldlet The re-sulting set of capture geometry constructs a 360 degspherical view from the current viewpoint

When displayed the spherical worldletrsquos geometrylooks like a bubble cut out of the VE A close yon clipplane keeps the bubble small ensuring that it capturesonly landmark features in the immediate neighborhoodand not the entire virtual world

53 Worldlet Capture in OpenGL

To take advantage of the rendering speed offeredby the accelerated 3-D graphics pipeline on high-speedworkstations we implemented worldlet display and cap-ture using OpenInventor (Wernecke 1994) andOpenGL graphics libraries from Silicon Graphics Scenegraph construction and display traversal is managed byOpenInventor To capture worldlet geometry theVRML browser places the pipeline into feedback mode

Figure 6 World shapes are transformed to screen space via the

downward OpenGL graphics pipeline on the left Worldlet shapes are

inverse transformed from window coordinates to object coordinates via

the upward pipeline in the middle of the gure The intermediary

coordinate systems are named on the right

572 PRESENCE VOLUME 10 NUMBER 6

prior to a capture traversal and returns it to renderingmode following traversal

While in feedback mode the OpenGL pipeline divertsall transformed clipped and culled coordinates into abuffer provided by the browser Upon completion of acapture traversal no rasterization has taken place andthe feedback buffer contains the extracted geometry Byparsing the feedback buffer the VRML browser recon-structs worldlet geometry applying appropriate inversetransforms

OpenGL feedback buffer information includes shapecoordinates colors and texture coordinates but it doesnot include an indication of which texture image to usefor which bit of geometry To capture this additionalinformation the VRML browser uses OpenGLrsquos pass-through features to pass custom ags down through thepipeline during traversal To prepare these pass-through ags the browser augments the world scenegraph prior to traversal assigning a unique identier toeach texture image During a capture traversal eachtime a texture image is encountered the associatedidentier is passed down through the pipeline and intothe feedback buffer along with shape coordinates col-ors and texture coordinates During parsing of thefeedback buffer these texture identiers enable worldletgeometry reconstruction to apply the correct textureimages to the correct shapes

6 Methods

The psychological experiment was designed to testwhether guidebooks of worldlets are better for waynd-ing than guidebooks of text and images and if so why

61 General Description

A standard within-subject randomized designwas used Training began by reading each subject a se-ries of instructions and teaching the subject in a prac-tice virtual city (three blocks by three blocks) how tooperate the user interface Subjects were allowed tospend unlimited time in this city becoming familiar withthe controls and the online guidebooks

Each subject was then asked to nd his or her wayvia one landmark to the goal (a brightly colored kiosk)in three practice cities each ve blocks by ve blocks Inone city the guidebook represented each landmark witha paragraph of text in another city the landmarks wererepresented as 2-D thumbnail images and in a third citythe landmarks were represented as spherical worldletsThe guidebooks contained only landmark informationand subjects were given no procedural or survey infor-mation Although the image and worldlet guidebookhad the same appearance as shown in gure 4 theworldlets in the worldlet guidebook could be exploredand manipulated

After completing the practice tasks the experimentbegan and subjects were asked to nd their way to thegoal kiosk via eight ordered landmarks in three virtualcities ten blocks by ten blocks Performance in thesethree cities was timed and the subjectrsquos location orien-tation and actions were recorded at 1 sec intervalsSubjects visited the test cities in a random order tocounterbalance possible carryover effects and each ofthe cities and landmarks were different and unique

Before the computer portion of the experiment sub-jects completed a brief questionnaire on their prior ex-perience with computers After the experiment subjectsanswered written questions about the strategies theyused while nding their way and how difcult theyfound the task using each of the guidebooks Subjectsalso completed spatial and verbal neuropsychologicaltests during a separate session

62 Constants

The major independent variable in this experimentwas the type of guidebook used (text image or world-let) Parameters concerned with city design and land-mark layout as listed in table 1 were held constant

63 City Design

Our primary goal in designing the virtual citieswas to preserve as many elements of a real city as possi-ble Toward this goal we developed a pseudo-randomcity generator that produced cities based on a regular

Elvins et al 573

street grid with pavement roads and sidewalks betweenthe blocks Each block contained up to twenty build-ings side by side around the blockrsquos perimeter Using acache of 250 building designs buildings were randomlyselected and placed on city blocks Many non-landmarkbuildings were repeated between and within the citiesRendering requirements were reduced by programmingbuildings to be invisible when farther than three blocksfrom the current viewpoint To prevent subjects fromseeing buildings appearing and disappearing in the dis-tance we added fog to the environment

Cities contained parks parking lots and other civicfeatures but did not contain cars or street signs Land-marks were placed so that a subject could not short-cutthe route and would never have to make an exhaustivesearch as long as they carefully looked down each possi-ble path for familiar landmarks and landmark contextthey could always determine in which direction to pro-ceed No landmarks were located within one block ofthe perimeter wall that surrounded each city

64 Radius of Information

The radius of information was the most difcultparameter to make constant across the three guide-

books Given that text 2-D images and worldlets aredifferent representations they do not convey an equiva-lent quantity and quality of landmark information Wedetermined while building the guidebooks for the pilotstudy that using a radius of 20 m produced the mostnearly equivalent set of guidebooks For the text guide-book this meant describing in words the landmark thebuilding or structure directly across the street thebuilding on each side and whether the landmark was ona corner Thus our textual description of the landmarkshown in gure 3 read

Citgo Gas Station Citgorsquos red gas pumps are at acorner location Next door is the Cafe 360 in a beigecolored building and across the street is the LumbiniRestaurant in a red brick building

For the image guidebook a 20 m radius meant posi-tioning the camera across the street from the landmarkso that the captured thumbnail image would contain thebuildings or structures on the left and right of the land-mark For the worldlet guidebook setting the informa-tion radius simply meant setting the yon clip plane tocutaway 3-D shapes farther away than 20 m while cap-turing a spherical worldlet Image and worldlet camera

Table 1 Although the Arrangement of Buildings Landmarks and Routes Were Different in Each City Their Basic Structurein Terms of City Design and Landmark Layout was Kept Constant to Allow for Controlled Comparison Subjects Movedat a Fixed Speed in all Cities

Parameter Constant value Comments

City dimensions ten by ten blocks 100 total blocksBuilding lots per block twenty on perimeter of block building footprint variesDistance from start to goal 1200 meters past 100 12-meter-wide lotsLandmarks per city nine including goal kiosk ordered in guidebooksDistance between landmarks 15 blocks 6 025 blockTravel speed ve meterssecond slow driving speedTypes of landmarks eight one of each kind in each cityTurns required to reach goal ve left and right turnsRadius of information approximately twenty meters see city design sectionLandmark positions three on corner ve midblock goal is midblockGoal building yellowred striped kiosk six-sided viewpoint invariant

574 PRESENCE VOLUME 10 NUMBER 6

locations were selected so as not to suggest a directionof approach to the landmark

65 Hardware and Software Design

The experiment was conducted using a standard19 in monitor Stereo glasses were not used A maincity window displayed the city in the waynding VRMLbrowser Subjects used keyboard arrow keys to moveforward and backward a xed distance on each keypress and to turn left and right by a xed angle on eachkey press In the worldlet guidebook subjects used thearrow keys in the same manner to move around in theworldlet Collision detection prevented subjects frompassing through buildings and other objects in the city

Subjects pushed a Start button before beginning aLandmark button each time a landmark was reachedand a Stop button upon reaching the goal kiosk Sub-jects pushed a Guidebook button to display the guide-book Subjects practiced operating all of these buttonsin practice cities prior to beginning the experiment Thewindows were designed so that when the guidebookwas open (shown in gure 4) the main city window(shown in gure 7) was completely occluded and nomovement could be made

66 Lessons Learned from the PilotStudy

Our pilot study design and results are described byElvins et al (1997) From the pilot study we learnedthat landmarks cannot be more than two blocks apartbecause subjects frequently resort to an exhaustivesearch of the neighborhood We also learned that somesubjects will take much longer than other subjects tocomplete the task and that the task difculty and lengthmust be designed while keeping the slower wayndersin mind

67 Subject Instructions

The following is a subset of the instructions readto the subjects Instructions 3 and 4 were repeated be-fore beginning each city

1 The guidebook contains a series of landmarks thatyou probably want to follow since traveling pastthese landmarks will lead you to the goal kiosk

2 When you are at a landmark you may have towalk as far as one or two blocks before you will beable to see the next landmark

Figure 7 The main city window

Elvins et al 575

3 Your task is to locate the goal kiosk as quickly aspossible Go as fast as you can but be careful tonot get lost

4 Remember you can look at the guidebook as of-ten as you like

68 Subjects

Subjects were primarily University of CaliforniandashSan Diego students from a number of major depart-ments Ages ranged from 18 to 36 with a mean age of24 Fourteen men and twelve women completed theexperiment Results from four other subjects had to beexcluded from our data because the subjects becamelost in one or more of the cities All subjects were com-puter literate but with varying degrees of experiencewith VEs

69 Neuropsychological Tests

To provide an independent measure of a subset ofthe subjectsrsquo cognitive abilities subjects were tested in aseparate session using two standardized subtests fromthe Wechsler Adult Intelligent Scale-Revised (WAIS-R)(Wechsler 1955) The Vocabulary subtest comprisingpart of the WAIS-R Verbal subtests was used to assessthe subjectrsquos verbal ability Performance was scored ac-cording to standardized procedures based on the sub-jectrsquos degree of familiarity with a list of up to 35 wordsand their elegance of expression The second test wasthe Block Design subtest comprising part of theWAIS-R Performance subtests which was used to assessthe subjectsrsquo constructional-spatial ability and was ad-ministered in the standardized manner

610 Analysis

A one-way repeated measures analysis of variance(ANOVA) was performed for each of the dependentvariables and the overall times and distances The with-in-subjects variable was the landmark description typewith three levels text image and worldlet Post hocanalyses were completed using the Scheffe test for dif-ferences in sample means Correlation analysis was per-

formed across the scores of the neuropsychological testsand the scores of the waynding tests We adopted asignicance level of 005 on all analyses unless otherwisenoted

7 Results

Our primary result was that people spend moretime consulting the worldlet guidebook than the otherguidebooks but that they require less time to reach thegoal when using the worldlet guidebook The followingspecic ndings tell us more about the component ac-tivities that benetted from worldlet use

71 Time and Distance Measures

Figures 8 through 11 show time and distancemeasures in which the independent variable (type ofguidebook) is displayed as white gray and black barsalong with the standard error of the mean (SEM) Fig-ure 8 shows the mean overall time and distance requiredto reach the goal kiosk and gure 9 breaks the overalltime into its components Figure 10 gives the forwardand backtracking distance components Figure 11 showsthe mean number of times that each type of guidebook

Figure 8 (left) Mean total time The overall time spent traveling to

the goal was signicantly less using the worldlet guidebook than when

using the text (F(1 24) 5 989 p 001) or image (F(1 24) 5

463 p 001) guidebooks (right) Mean total distance The overall

distance traveled using the worldlet guidebook was signicantly less

than the distance traveled using the text (F(1 24) 5 1141 p

001) or image (F(1 24) 5 893 p 001) guidebooks

576 PRESENCE VOLUME 10 NUMBER 6

was opened Figure 12 diagrams the subjectsrsquo scores onthe Wechsler neuropsychological tests

Analysis of variance followed by Scheffe post hoc testswere performed for the different measures Table 2summarizes these results

72 Street Plots

Figure 13 shows plots of the streets traversed by the26 subjects in the three test cities The thickness andbrightness of each street line indicate the total number oftimes that the street was traversed For example widerbrighter streets indicate heavily used avenues Backtrackingcan increase the number of street traversals beyond theexpected minimum of 26 The starting point and goalkiosk are indicated on each plot and landmarks enroute tothe goal are marked with a star (q )

Figure 10 The mean distance moving forward while using the

worldlet guidebook is signicantly less than the forward distance

covered while using the text or image guidebook Backtracking

distance is almost nonexistent for the worldlet guidebook task

Figure 9 Turning looking for landmark is time spent standing in

one location Turning changing forward direction is time spent

changing forward direction Consulting guidebook is time spent with

the guidebook on screen Moving forward is time spent traveling over

new ground and Backtracking is time spent retreating Note that the

time spent consulting the worldlet guidebook is signicantly greater

than the time spent consulting the text or image guidebook Also note

that the time spent backtracking is near zero when using the worldlet

guidebook

Figure 11 Subjects opened the text and image guidebooks an

average of fteen to sixteen times while seeking the eight landmarks

enroute to the goal kiosk The worldlet guidebook was opened eleven

times on average

Figure 12 Z-transformed spatial and verbal scores from the

neuropsychological tests The mean raw scores were 45 and 55 for

the spatial and verbal tests respectively

Elvins et al 577

73 Subjective Measures

As part of the written questionnaire administeredafter the experiment subjects were asked how difcultthe search task was using each of the different types ofguidebooks Although responses were scattered a cleartrend in favor of the worldlet guidebook was evident asshown in table 3

74 Correlation Analysis

One surprising nding was a high inverse corre-lation ( 2 077 p 5 001) between spatial skills andthe time spent consulting the text guidebook Sub-jects with better spatial ability seemed able to form amental picture of a landmark based on textual de-scription more quickly than did subjects with poorer

spatial ability This is not the case for image andworldlet guidebooks however

We expected to nd a negative correlation betweenspatial skills and overall travel time in the image andworldlet guidebook cities The data did not supportthis nor did they support another of our hypothesesthat there would be a positive correlation between ver-bal skills and the performance in the text guidebookcity Verbal skills made no difference in overall waynd-ing time

75 Observations on Subject Strategies

Many subjects said that the tasks were harderthan they had imagined and afterwards several saidthat they felt tired No subjects complained of mo-

Table 2 Our ANOVA Indicated Signicant Differences between All the Sample Means and Post Hoc Analysis Was Warranted

ANOVA

Post hoc comparisons (Scheffe F-test)

text vs image text vs worldlet image vs worldlet

TimeConsulting guidebook 5559 p 00001 1966 p 0001 884 p 001 5488 p 0011Looking for landmark 2161 p 00001 047 1869 p 0001 1325 p 001Changing heading 1736 p 00001 002 1353 p 001 1249 p 001Moving forward 1602 p 00001 0003 1220 p 001 1183 p 001Backtracking 955 p 00003 022 826 p 001 586 p 001Overall 1034 p 00002 098 989 p 001 463 p 001

DistanceMoving forward 1533 p 00001 005 1223 p 001 1072 p 001Backtracking 1019 p 00002 023 884 p 001 621 p 001Overall 1365 p 00001 015 1141 p 001 893 p 001

GuidebookOpened 1527 p 00001 104 759 p 001 1426 p 0001

Post hoc (Scheffe) analysis indicated for all dependent variables signicant differences in the sample means of the textversus worldlet performance and image versus worldlet performance Performance using the text versus imageguidebooks was not signicantly different except that the time spent consulting the text guidebook was signicantlygreater than the time spent consulting the image guidebook In each table cell the rst number is the F value and thesecond number is the p value Asterisk indicates signicance p 005 Time and distance measures are explained in thecaptions for gures 8 through 11

578 PRESENCE VOLUME 10 NUMBER 6

tion sickness Most subjects required 70 min or moreto nish all aspects of the computer-based part of theexperiment

We observed subjects to use three general types ofstrategies while deciding what direction to go from astreet intersection

c Look down each street from many viewpointswithin the intersection until a landmark or land-mark context is seen

c Look down each street and if nothing looks famil-iar search exhaustively in each direction until thenext landmark is found

c Look down each street and if nothing looks famil-iar try going one-half block in each direction andlook again

Subjects using the rst strategy seemed to nd theirway more quickly and with less backtracking than didsubjects who used the second and third strategies An-other strategy described by some subjects on their writ-ten questionnaire involved viewing a landmark in theworldlet guidebook from many distant viewpoints in thehopes that one of the viewpoints would be similar to thedirection from which they would actually approach the

Figure 13 In these street plots line thickness and brightness

indicates the number of times that a street was traversed In top and

middle the 26 subjects traversed some streets 47 times indicating

backtracking Top and middle also exhibit circuitous trails In bottom

very little backtracking occurred and circuitous trails are minimal

Table 3 Subjects Were Asked to Rate the Difculty of theSearch Task Using the Three Different Types of Guidebooks

Text Image Worldlet

Very easy 0 2 11Easy 10 8 9Doable 10 8 6Difcult 5 7 0Very difcult 1 1 0

Median Doable Doable Easy

Elvins et al 579

landmark in the city We observed that some subjectsoften forgot what it was they were looking for and sohad to open the guidebook momentarily to remindthemselves

Not surprisingly landmarks with prominent signage(such as ldquoBobrsquos Groceryrdquo) and landmarks painted withmore highly saturated colors seemed easier to nd thanother types of landmarks Also after passing all thelandmarks several subjects exclaimed that they had lostall sense of direction within the world

A nal unexpected result were the many similarities be-tween the text and image guidebooks The text and imageguidebooks were only signicantly different in the timesubjects spent consulting them Considering that text andimages represent landmarks in very different ways we ex-pected differences in time and distance measures

76 Major Findings

In interpreting these ndings it is important toremember that the subjects performed the wayndingtasks using only the landmark knowledge gained bystudying the guidebooks No procedural or survey(map) information was provided

c As we predicted the total time spent traveling tothe goal was signicantly less using the worldletguidebook than when using the text or imageguidebooks

c Subjects spent signicantly more time studying theworldlet guidebook than they did the text or imageguidebook

c Increased worldlet study time was more than com-pensated for by a reduction in the time required toreach the goal Using the worldlet guidebook sub-jects traveled an average of 55 faster than whenusing the text guidebook and 38 faster than whenusing the image guidebook

c Increased worldlet study time also allowed subjectsto move more directly to the goal and reducedbacktracking time to almost zero This is showndiagramatically in gure 13 in which gures 13aand 13b show brighter wider streets than in gure

13c Figure 13c also exhibits fewer detours and lesswandering than do gures 13a and 13b

We postulate that compared to the landmark knowl-edge gained from text and image guidebooks the im-proved landmark knowledge built while consulting theworldlet guidebook enabled subjects to refer less fre-quently to the guidebook This postulation is furthersupported by examining worldlet guidebook results

c Subjects spent signicantly more time studying theworldlet guidebook than they did the other types ofguidebooks

c When using the worldlet guidebook subjects trav-eled faster more directly and with more skill tothe goal kiosk

c Subjectsrsquo median rating for the difculty of thesearch task was Easy when using the worldlet guide-book

8 Conclusion

The waynding literature shows that landmarksare critical in navigating an environment whether it isreal or virtual While building landmark knowledgetravelers characterize a landmark representation by its3-D shape size texture context and so forth In awaynding task textual or image guidebook landmarkdescriptions fail to engage the full range of 3-D land-mark characteristics recognized and used by travelers tond their way

We know of no studies that have considered the com-parative advantages of different types of guidebooks forwaynding Intuitively one would predict that 3-Dthumbnails make better guidebooks for 3-D worldsthan do text and 2-D thumbnail images and indeedthis is what we found We wanted however to identifythe component activities in using guidebooks that mostbenetted from 3-D elements Perhaps of greatest inter-est was our nding that worldlets enabled subjects tond their way in complex environments with almost nobacktracking By spending more time studying theworldlets in their guidebooks they spent less time look-ing around reorienting themselves conducting exhaus-

580 PRESENCE VOLUME 10 NUMBER 6

tive searches and performing similar time-wasting activ-ities We must be cautious however in what weconclude because worldlets contain more informationthan do text or 2-D thumbnail images To control forthis difference in information content more experi-ments are necessary

9 Future Work

We would like to investigate the comparative ad-vantages of worldlet guidebooks for waynding tasks incorresponding virtual and real world environments(both urban and rural) We are also considering an ex-periment to evaluate tandem use of traditional 2-Dmaps and landmark guidebooks Requiring effectivemap and compass use in a natural environment thesport of orienteering (Kjellstrom 1976) might benetfrom worldlets By encapsulating the region around theagged control points into a worldlet orienteers couldgain rst-person experience with natural landmarks suchas foliage rocks gullies shadows and so on beforebeginning the search

Acknowledgments

Our appreciation goes to Planet 9 Studios (wwwplanet9com)for the use of their city building textures Thanks to JohnMoreland and Mike Bailey for technical assistance on thisproject and thanks to Suzi Feeney for collaborating in thepilot study design UCSD librarian Debbie Cox and her staffwere invaluable in locating references Thanks also to JohnHelly and Reagan Moore for their support San Diego Super-computer Center is funded by NSF the University of Califor-nia the State of California and industrial partners

References

Appleyard D A (1976) Planning a Pluralistic City Con-icting Realities in Ciudad Guayana Cambridge MAMIT Press

Art1 Com (1997) The cybercity Berlin project [on-line]Available httpwwwartcomdeprojects

Chen S E (1995) QuickTime VRmdashAn image-based ap-proach to VE navigation Proceedings of SIGGRAPH 9529 ndash38

Darken R P amp Sibert J L (1996) Waynding strategiesand behaviors in large virtual worlds Proceedings of CHI 96142ndash149

Downs R J amp Stea D (1973) Cognitive maps and spatialbehavior In R J Downs amp D Stea (Eds) Image and En-vironment Chicago Aldine Publishing Company 8 ndash26

Elvins T T (1997) Waynding in layered informationworlds Proceedings of the CODATA Euro-American Work-shop on Data and Information Visualization 6 ndash7

Elvins T T Nadeau D R amp Kirsh D (1997) World-letsmdash3D thumbnails for waynding in virtual environmentsProceedings of the ACM User Interface Software and Technol-ogy Symposium 21ndash30

Elvins T T (1998) WorldletsmdashA novel 3D thumbnail af-fordance for waynding in large virtual environmentsUnpublished doctoral dissertation Department of Elec-trical and Computer Engineering University of Califor-niandashSan Diego

Elvins T T Nadeau D R Kirsh D amp Schul R (1998)Worldletsmdash3D thumbnails for 3D browsing Proceedings ofthe ACM CHI Conference 21ndash30

Foley J van Dam A Feiner S amp Hughes J (1990) Com-puter Graphics Principles and Practice Reading MA Addi-son-Wesley

Gillner S amp Mallot H A (1998) Navigation and acquisi-tion of spatial knowledge in a virtual maze Journal of Cog-nitive Neuroscience 10(4) 445ndash 463

Goldin S E amp Thorndyke P W (1982) Simulating naviga-tion for spatial knowledge acquisition Human Factors24(4) 457ndash 471

ISO (1997) The Virtual Reality Modeling Language Inter-national Standard ISOIEC 14772-11997 [on-line] Avail-able httpwwwvrmlorgSpecicationsVRML97

Kjellstrom B (1976) Be Expert with Map and Compass TheComplete Orienteering Handbook New York Scribner Pub-lishers

Liggett R Friedman S amp Jepson W (1995) Interactivedesigndecision making in a virtual urban world visual sim-ulation and GIS Proceedings of the 1995 ESRI User Confer-ence Environmental Systems Research [on-line] Availablehttpwwwesricombasecommonuserconfproc95

Lynch K (1960) The Image of the City Cambridge MAMIT Press

Elvins et al 581

c What aspects of the waynding process are mostenhanced by the use of worldlets

In the experiment subjects were placed at startinglocations in three virtual cities and asked to nd theirway to goal locations via a series of landmarks In onecity the landmarks were described with text in anothercity the landmarks were described using 2-D thumbnailimages and in yet another city the landmarks were de-scribed using manipulable worldlets as shown in gure3 We hypothesized that compared to text and thumb-nail images worldlets would facilitate better landmarkknowledge and expedite waynding

2 Way nding

Waynding is ldquoa term that can refer to a rather nar-row concern That is how well people are able to ndtheir way to their particular destination without delay orundue anxietyrdquo (Peponis Zimring amp Choi 1990 p561) Whereas navigating is most often dened as theprocess of following a course by making directed move-ments waynding is the critical process of determining

the strategy direction and course needed to reach adesired destination Travelers nd their way using sur-vey procedural and landmark knowledge (Goldin ampThorndyke 1982 Peponis et al 1990) Each type ofknowledge helps the traveler construct a cognitive mapof a region and thereafter nd their way using that map(Appleyard 1976 Downs amp Stea 1973 Lynch 1960Passini 1992)

Survey knowledge provides a map-like birdrsquos-eyeview of a region and contains spatial information includ-ing locations orientations and sizes of regional fea-tures Procedural knowledge characterizes a region bymemorized sequences of actions that construct routes todesired destinations Landmark knowledge records thevisual features of landmarks including their 3-D shapesize texture and so forth (Appleyard 1976 Goldin amp

Figure 1 A Web page linking to alternative destinations within a

virtual world On this page the destinations are represented with text

Figure 2 On this Web page the destinations from Figure 1 are

represented as 2-D thumbnail images

566 PRESENCE VOLUME 10 NUMBER 6

Thorndyke 1982) For a structure to be a landmark itmust have high imagability that is it must be distinctiveand memorable (Lynch 1960)

Landmarks are the subject of landmark knowledgebut they also play a part in survey and proceduralknowledge In survey knowledge landmarks provideregional anchors with which to calibrate distancesand directions In procedural knowledge landmarksmark decision points along a route helping in therecall of procedures to get to and from destinationsof interest

Whereas a travelerrsquos landmark knowledge charac-terizes a destination by its 3-D shape size textureand so forth traditional virtual environment (VE)browser menus and landmark guidebooks character-ize destinations only by textual descriptions or im-ages This representation mismatch reduces the effec-tiveness of destination menus and guidebooksUnable to engage their memory of 3-D landmarksto recognize destinations of interest travelers mayresort to a naive exhaustive search to nd a desiredlandmark

Figure 3 Construction of a worldlet A gas station landmark in a virtual city (top left) viewed from a

viewpoint in the street (top right) Viewed from above the gas station is upper left and a region

around the previous viewpoint is highlighted (bottom left) The gas station is captured within a spherical

worldlet and (bottom right) is viewed from street level within the worldlet

Elvins et al 567

21 Related Work

Although social and engineering scientists havestudied human waynding in the real world for manyyears (Appleyard 1976 Downs amp Stea 1973 Goldinamp Thorndyke 1982) only recently have computer sci-entists investigated this subject Satalich (1995) led sub-jects on a tour of a virtual building and then tested thesubjectsrsquo ability to nd their way to places seen duringthe tour Darken and Sibert (1996) applied Lynchrsquos(1960) city structure theories to large sparsely featuredVEs and then compared the subjectsrsquo ability to ndtheir way to target destinations with and without theadded city structure Witmer Bailey Knerr and Parsons(1996) studied route learning and found that subjectswho rehearsed a route in a virtual building learned theroute better than did subjects who rehearsed the routeverbally Gillner and Mallot (1997) found that subjectspieced together survey knowledge when repeatedly re-quired to nd shortest paths between a dozen unfamil-iar locations in a VE In studying VEs for real-worldtask training Waller Hunt and Knapp (1998) foundthat long exposure fostered good spatial knowledge un-der several VE delity conditions Ruddle Payne andJones (1998) studied path complexity and a compasscondition as factors in VE route learning and foundthat beyond two turns subjects were quickly disori-ented with or without a compass Although each ofthese studies recognized the value of landmark knowl-edge for waynding none studied the value of 3-Dlandmarks in familiarizing subjects with an environmentbefore entering it This is a primary goal of our study

22 Way nding Affordances

Three-dimensional models of large cities are be-coming available for exploration via the Web(Art 1 Com 1997 Liggett Friedman and Jepson1995 University College London 1997) As discussedpreviously travelers visiting such large VEs for the rsttime are easily disoriented may have difculty identify-ing a place upon arrival and may not be able to ndtheir way back to a place just visited (Downs amp Stea1973 Darken amp Sibert 1996) The following subsec-

tions describe affordances that travelers currently use toscrutinize potential landmark and VE destinations

c Textual descriptions are the dominant method usedto represent VE landmarks in world- and viewpoint-selection user interfaces HTML pages for instanceoften provide lists of available Web-based worlds(such as those authored in the Virtual Reality Mod-eling Language (VRML) (ISO 1997)) each onecharacterized by a URL a VE name andor a briefdescription Within VRML worlds textual descrip-tions characterize viewpoints and describe destina-tions associated with clickable anchor shapes

c Icons thumbnail images and image maps providecommon visual waynding aids In a 3-D contextgames often provide ldquojump gatesrdquo onto which im-ages of remote destinations are texture mappedStepping through such a gate jumps the traveler tothe destination landmark depicted on the gate

c Image mosaics group together multiple capturedimages into a traversable structure Applersquos Quick-Time VR for instance can use images capturedfrom multiple viewing angles at the same viewingposition (Chen 1995) By ordering images within atraveler-centered cylindrical structure QuickTimeVR can provide a traveler the ability to look in anydirection through automatic selection of an appro-priate image from the structure By chaining multi-ple mosaic structures together the content authorcan create a walkthrough path that hops from view-point to viewpoint Similar image mosaics can beused to create zoom paths pan paths and so forth

c Miniature Worlds and Maps are available in most3-D environment browsers and computer gamesand they enable the traveler to zoom out and viewthe world in miniature and to thereby gain surveyknowledge Pausch Burnette Brockway andWeiblen (1995) extend this notion by creating aworld in miniature (WIM) embedded within themain world The miniature world duplicates all ele-ments of the main world and adds an icon denotingthe travelerrsquos position and orientation Held withinthe travelerrsquos virtual hand the traveler can reachinto the miniature and reposition world content

568 PRESENCE VOLUME 10 NUMBER 6

andor themselves Simultaneously the outer mainworld is updated to match the altered miniatureautomatically adjusting the positions of shapes orthe traveler Similarly 2-D and 3-D maps are fre-quently found as navigation aids within VEs Forinstance 3-D games often provide a 3-D reduced-detail map in which an icon denotes the playerrsquoslocation Such maps can be panned zoomed androtated to provide alternate viewpoints similar tothat which is possible with miniature worlds

None of these waynding affordances offer the trav-eler faithful 3-D landmark representations or a rst-person experience with potential destinations Wickensand Baker (1995) argue that VE search tasks are dif-cult because ldquooften the object of the search is speciedin a format different from and more abstract than itsrendering in the VErdquo We address this problem by rep-resenting the object of the search in the same 3-D for-mat as the VE

3 What are Worldlets

In previous papers we described our ongoingworldlet waynding research (Elvins 1997 ElvinsNadeau amp Kirsh 1997 Elvins 1998 Elvins NadeauKirsh amp Schul 1998) Similar to a 3-D photograph aworldlet is associated with a viewing position and orien-tation and captures a snapshot of the 3-D shapes (usu-ally landmarks) that fall within the viewpointrsquos viewingvolume Worldlets can be explored and manipulated inthe same way that virtual worlds can be explored andmanipulated Incorporated into a VRML browser aguidebook or list of worldlets enables a world traveler tointeractively view landmarks and landmark context frommultiple viewpoints Seeing the landmarks presented inworldlets helps a traveler select destinations of interestand nd their way in unfamiliar worlds

In typical use the worldletrsquos viewpoint is aimed at animportant landmark and the worldletrsquos captured shapesreconstruct that landmark and its associated contextWhen viewed within an interactive 3-D browser a

worldlet provides a manipulable 3-D thumbnail repre-sentation of the landmark

Clicking on a worldlet in a landmark menu mightjump the traveler to the landmark represented by theworldlet The landmark could be at a viewpoint in thecurrent world or in another world A menu of worldletssuch as that shown in gure 4 might resemble land-mark history lists bookmarks or travel guidebooksTravelers are free to study the landmark in the worldletfrom any viewpoint before deciding whether to gothere

We have developed two types of worldlets

c A frustum worldlet contains shapes within a stan-dard pie-shaped viewing frustum positioned andoriented based upon a selected viewpoint Whenviewed a frustum worldlet looks like a pie-shapedfragment clipped from the larger main world

c A spherical worldlet contains shapes within a spheri-cal viewing bubble positioned at a selected view-point with a 360 deg eld of view When dis-played a spherical worldlet looks like a ball-shapedworld fragment similar to a snowglobe knick knackdepicting a favorite attraction

Figure 4 The navigation browserrsquos viewpoint selection window and

the waynding browserrsquos worldlet guidebook The ldquoGo tordquo button was

removed in the waynding browser

Elvins et al 569

For both worldlet types hither and yon clipping planesrestrict the extent of the worldlet ensuring that theworldlet contains a manageable subset of the largermain world Worldlet shape content is shaded and tex-tured to match the corresponding shapes in the mainworld Although the main world may have content thatchanges over time the captured worldlet remains staticrecording the content of the world at the time theworldlet was captured

Figure 3 depicts the process of constructing a spheri-cal worldlet from a fragment of a virtual city surround-ing a central viewpoint position Once captured world-lets can be browsed manipulated and explored in anumber of ways

4 Worldlets in Browsers

We have incorporated worldlets into the user in-terface for two experimental VRML browsers a naviga-tion browser and a waynding browser The navigationbrowser provides a ldquoGo tordquo feature that enables travel-ers to study select and y to world viewpoints in thecurrent world and to study select and load previouslyvisited worlds on the browserrsquos history list Guidebooksof worldlets in the waynding browser allow travelers tostudy 3-D landmark guidebooks before and duringtravel in an unfamiliar VE No go-to button is presentin this browser in which travelers manually move them-selves about based on the landmark knowledge gleanedduring study

41 Navigation Browser

Traditional VRML browsers provide a viewpointmenu that offers a choice of viewpoints each denotedby a brief textual description We have extended thisstandard feature to provide three experimental view-point selection interfaces each using worldlets All threepresent a set of worldlets one for each author-selectedviewpoint in the world A ldquoGo tordquo button ies the mainwindowrsquos viewpoint to that of the currently selectedworldlet The browser also supports on-the-y captureof worldlets from the travelerrsquos current viewpoint

c The viewpoint list window provides a list of world-lets beside a worldlet viewer Selection of a worldletfrom the list displays the worldlet in the viewerwhere it may be interactively panned zoomed androtated

c The viewpoint guidebook window presents a grid ofworldlet viewers arranged to form a guidebookphoto-album page Buttons on the window advancethe guidebook forward or back a page at a timeEach worldlet on the page can be interactively ex-amined Figure 4 shows the viewpoint guidebookwindow

c The viewpoint overlay window enables the travelerto select a worldlet from a list and overlay it atopthe main window highlighted in green This world-let overlay provides a clear indication of the world-letrsquos viewpoint position and orientation along withthe portion of the world captured within thatworldlet Figure 3b was generated using this overlaytechnique

Traditional VRML browsers provide a history list ofrecently visited worlds each denoted by its title orURL We have extended this standard feature to pro-vide two world selection interfaces each using world-lets In each world selection window a ldquoGo tordquo buttonloads into the main window the world associated withthe currently selected worldlet

c The world list window provides a list of worldletsbeside an interactive worldlet viewer similar to theviewpoint list window discussed earlier One world-let is available for each world on the browserrsquos his-tory list

c The world guidebook window uses the same guide-book photo-album layout used for the viewpointguidebook window discussed earlier One worldletis available for each world on the history list

A ldquoSave asrdquo feature of the VRML browser enables thetraveler to save a worldlet to a VRML le Using a col-lection of saved worldlets a world author can create aVRML world of worldlets Such a world acts like a 3-Ddestination index similar to a shelf full of snowglobeknick knacks When cast as a VRML anchor shape a

570 PRESENCE VOLUME 10 NUMBER 6

worldlet provides a 3-D button that when clicked onloads the associated world into the travelerrsquos browser

Figure 5 shows a world of worldlets Figure 5a shows

a close-up view of a world ldquodoorwayrdquo and a niche thatcontains a worldlet illustrating a viewpoint in thatworld Figure 5b shows a wider view of the same worldand multiple such doorways

42 Way nding Browser

The second browser was implemented specicallyfor the psychological experiment After manipulatingworldlets in a landmark guidebook to familiarize them-selves with the 3-D size shape color texture and con-text of the landmarks subjects in the experiment foundtheir way manually in the main window based upontheir landmark knowledge The landmark guidebookresembled the viewpoint selection window shown ingure 4 but it did not offer a ldquoGo tordquo button Thewaynding browser is described in detail in the methodssection

5 Implementation

Worldlets can be implemented in several waysThe implementation presented here is based on ahardware-accelerated algorithm and support algorithmsdeveloped specically for fast execution The VRMLbrowser used in this work maintains VE geometrywithin a tree-like scene graph Worldlets are also storedas scene graphs together with additional state informa-tion To capture a worldlet or to display a worldlet orVE the VRML browser traverses the associated scenegraph and feeds a 3-D graphics pipeline

51 Worldlet Capture in General

Any 3-D graphics pipeline can be roughly dividedinto two stages (i) transform clip and cull and (ii)rasterize (Foley van Dam Feiner amp Hughes 1990)The rst stage applies modeling viewing perspectiveand viewport transforms to map 3-D shapes to the 2-Dviewport Along the way shapes outside of the viewingfrustum are clipped away and backfaces are removedThe second stage uses 2-D shapes output by the rst

Figure 5 A world of worldlets that (top) associates a worldlet with

each doorway (bottom) in an environment containing multiple such

doorways Each doorway leads to a different world

Elvins et al 571

stage and draws the associated points lines and poly-gons on the screen

Worldlet capture taps into this 3-D graphics pipelineextracting the transformed shaded clipped and culledshape coordinates output by the rst stage prior to ras-terization in the second stage An extracted coordinatecontains x and y screen-space components a depth-buffer z-space component and the w coordinate Eachextracted coordinate has an associated RGB color andtexture coordinates which are computed by shadingand texture calculation phases in the rst pipeline stage

To create a worldlet these extracted coordinates areuntransformed to map them from viewport space backto world space as shown in gure 6 The inverses of the

viewport perspective viewing and modeling transformsare each applied Coordinate RGB colors and texturecoordinates are used to reconstruct 3-D worldlet geom-etry in a worldlet scene graph

Display of a worldlet passes this 3-D geometry backdown the graphics pipeline transforming clipping cull-ing and rasterizing the worldlet like any other 3-D con-tent

52 Frustum and Spherical Worldlets

A frustum worldlet is the result of capturing 3-Dgraphics pipeline output for a single traversal of thescene graph as viewed from the travelerrsquos current view-point The shape set extracted after the rst pipelinestage contains only those points lines and polygonsthat fall within the viewing frustum The worldlet con-structed by the browser from this geometry looks like apie-shaped slice cut out of the world

A spherical worldlet is the result of performing multi-ple frustum captures and combining the results TheVRML browser captures a spherical worldlet by spin-ning about a viewpoint position generating a set offrustum worldlets that each use a different viewing ori-entation Additional captures aimed straight up andstraight down complete the spherical worldlet The re-sulting set of capture geometry constructs a 360 degspherical view from the current viewpoint

When displayed the spherical worldletrsquos geometrylooks like a bubble cut out of the VE A close yon clipplane keeps the bubble small ensuring that it capturesonly landmark features in the immediate neighborhoodand not the entire virtual world

53 Worldlet Capture in OpenGL

To take advantage of the rendering speed offeredby the accelerated 3-D graphics pipeline on high-speedworkstations we implemented worldlet display and cap-ture using OpenInventor (Wernecke 1994) andOpenGL graphics libraries from Silicon Graphics Scenegraph construction and display traversal is managed byOpenInventor To capture worldlet geometry theVRML browser places the pipeline into feedback mode

Figure 6 World shapes are transformed to screen space via the

downward OpenGL graphics pipeline on the left Worldlet shapes are

inverse transformed from window coordinates to object coordinates via

the upward pipeline in the middle of the gure The intermediary

coordinate systems are named on the right

572 PRESENCE VOLUME 10 NUMBER 6

prior to a capture traversal and returns it to renderingmode following traversal

While in feedback mode the OpenGL pipeline divertsall transformed clipped and culled coordinates into abuffer provided by the browser Upon completion of acapture traversal no rasterization has taken place andthe feedback buffer contains the extracted geometry Byparsing the feedback buffer the VRML browser recon-structs worldlet geometry applying appropriate inversetransforms

OpenGL feedback buffer information includes shapecoordinates colors and texture coordinates but it doesnot include an indication of which texture image to usefor which bit of geometry To capture this additionalinformation the VRML browser uses OpenGLrsquos pass-through features to pass custom ags down through thepipeline during traversal To prepare these pass-through ags the browser augments the world scenegraph prior to traversal assigning a unique identier toeach texture image During a capture traversal eachtime a texture image is encountered the associatedidentier is passed down through the pipeline and intothe feedback buffer along with shape coordinates col-ors and texture coordinates During parsing of thefeedback buffer these texture identiers enable worldletgeometry reconstruction to apply the correct textureimages to the correct shapes

6 Methods

The psychological experiment was designed to testwhether guidebooks of worldlets are better for waynd-ing than guidebooks of text and images and if so why

61 General Description

A standard within-subject randomized designwas used Training began by reading each subject a se-ries of instructions and teaching the subject in a prac-tice virtual city (three blocks by three blocks) how tooperate the user interface Subjects were allowed tospend unlimited time in this city becoming familiar withthe controls and the online guidebooks

Each subject was then asked to nd his or her wayvia one landmark to the goal (a brightly colored kiosk)in three practice cities each ve blocks by ve blocks Inone city the guidebook represented each landmark witha paragraph of text in another city the landmarks wererepresented as 2-D thumbnail images and in a third citythe landmarks were represented as spherical worldletsThe guidebooks contained only landmark informationand subjects were given no procedural or survey infor-mation Although the image and worldlet guidebookhad the same appearance as shown in gure 4 theworldlets in the worldlet guidebook could be exploredand manipulated

After completing the practice tasks the experimentbegan and subjects were asked to nd their way to thegoal kiosk via eight ordered landmarks in three virtualcities ten blocks by ten blocks Performance in thesethree cities was timed and the subjectrsquos location orien-tation and actions were recorded at 1 sec intervalsSubjects visited the test cities in a random order tocounterbalance possible carryover effects and each ofthe cities and landmarks were different and unique

Before the computer portion of the experiment sub-jects completed a brief questionnaire on their prior ex-perience with computers After the experiment subjectsanswered written questions about the strategies theyused while nding their way and how difcult theyfound the task using each of the guidebooks Subjectsalso completed spatial and verbal neuropsychologicaltests during a separate session

62 Constants

The major independent variable in this experimentwas the type of guidebook used (text image or world-let) Parameters concerned with city design and land-mark layout as listed in table 1 were held constant

63 City Design

Our primary goal in designing the virtual citieswas to preserve as many elements of a real city as possi-ble Toward this goal we developed a pseudo-randomcity generator that produced cities based on a regular

Elvins et al 573

street grid with pavement roads and sidewalks betweenthe blocks Each block contained up to twenty build-ings side by side around the blockrsquos perimeter Using acache of 250 building designs buildings were randomlyselected and placed on city blocks Many non-landmarkbuildings were repeated between and within the citiesRendering requirements were reduced by programmingbuildings to be invisible when farther than three blocksfrom the current viewpoint To prevent subjects fromseeing buildings appearing and disappearing in the dis-tance we added fog to the environment

Cities contained parks parking lots and other civicfeatures but did not contain cars or street signs Land-marks were placed so that a subject could not short-cutthe route and would never have to make an exhaustivesearch as long as they carefully looked down each possi-ble path for familiar landmarks and landmark contextthey could always determine in which direction to pro-ceed No landmarks were located within one block ofthe perimeter wall that surrounded each city

64 Radius of Information

The radius of information was the most difcultparameter to make constant across the three guide-

books Given that text 2-D images and worldlets aredifferent representations they do not convey an equiva-lent quantity and quality of landmark information Wedetermined while building the guidebooks for the pilotstudy that using a radius of 20 m produced the mostnearly equivalent set of guidebooks For the text guide-book this meant describing in words the landmark thebuilding or structure directly across the street thebuilding on each side and whether the landmark was ona corner Thus our textual description of the landmarkshown in gure 3 read

Citgo Gas Station Citgorsquos red gas pumps are at acorner location Next door is the Cafe 360 in a beigecolored building and across the street is the LumbiniRestaurant in a red brick building

For the image guidebook a 20 m radius meant posi-tioning the camera across the street from the landmarkso that the captured thumbnail image would contain thebuildings or structures on the left and right of the land-mark For the worldlet guidebook setting the informa-tion radius simply meant setting the yon clip plane tocutaway 3-D shapes farther away than 20 m while cap-turing a spherical worldlet Image and worldlet camera

Table 1 Although the Arrangement of Buildings Landmarks and Routes Were Different in Each City Their Basic Structurein Terms of City Design and Landmark Layout was Kept Constant to Allow for Controlled Comparison Subjects Movedat a Fixed Speed in all Cities

Parameter Constant value Comments

City dimensions ten by ten blocks 100 total blocksBuilding lots per block twenty on perimeter of block building footprint variesDistance from start to goal 1200 meters past 100 12-meter-wide lotsLandmarks per city nine including goal kiosk ordered in guidebooksDistance between landmarks 15 blocks 6 025 blockTravel speed ve meterssecond slow driving speedTypes of landmarks eight one of each kind in each cityTurns required to reach goal ve left and right turnsRadius of information approximately twenty meters see city design sectionLandmark positions three on corner ve midblock goal is midblockGoal building yellowred striped kiosk six-sided viewpoint invariant

574 PRESENCE VOLUME 10 NUMBER 6

locations were selected so as not to suggest a directionof approach to the landmark

65 Hardware and Software Design

The experiment was conducted using a standard19 in monitor Stereo glasses were not used A maincity window displayed the city in the waynding VRMLbrowser Subjects used keyboard arrow keys to moveforward and backward a xed distance on each keypress and to turn left and right by a xed angle on eachkey press In the worldlet guidebook subjects used thearrow keys in the same manner to move around in theworldlet Collision detection prevented subjects frompassing through buildings and other objects in the city

Subjects pushed a Start button before beginning aLandmark button each time a landmark was reachedand a Stop button upon reaching the goal kiosk Sub-jects pushed a Guidebook button to display the guide-book Subjects practiced operating all of these buttonsin practice cities prior to beginning the experiment Thewindows were designed so that when the guidebookwas open (shown in gure 4) the main city window(shown in gure 7) was completely occluded and nomovement could be made

66 Lessons Learned from the PilotStudy

Our pilot study design and results are described byElvins et al (1997) From the pilot study we learnedthat landmarks cannot be more than two blocks apartbecause subjects frequently resort to an exhaustivesearch of the neighborhood We also learned that somesubjects will take much longer than other subjects tocomplete the task and that the task difculty and lengthmust be designed while keeping the slower wayndersin mind

67 Subject Instructions

The following is a subset of the instructions readto the subjects Instructions 3 and 4 were repeated be-fore beginning each city

1 The guidebook contains a series of landmarks thatyou probably want to follow since traveling pastthese landmarks will lead you to the goal kiosk

2 When you are at a landmark you may have towalk as far as one or two blocks before you will beable to see the next landmark

Figure 7 The main city window

Elvins et al 575

3 Your task is to locate the goal kiosk as quickly aspossible Go as fast as you can but be careful tonot get lost

4 Remember you can look at the guidebook as of-ten as you like

68 Subjects

Subjects were primarily University of CaliforniandashSan Diego students from a number of major depart-ments Ages ranged from 18 to 36 with a mean age of24 Fourteen men and twelve women completed theexperiment Results from four other subjects had to beexcluded from our data because the subjects becamelost in one or more of the cities All subjects were com-puter literate but with varying degrees of experiencewith VEs

69 Neuropsychological Tests

To provide an independent measure of a subset ofthe subjectsrsquo cognitive abilities subjects were tested in aseparate session using two standardized subtests fromthe Wechsler Adult Intelligent Scale-Revised (WAIS-R)(Wechsler 1955) The Vocabulary subtest comprisingpart of the WAIS-R Verbal subtests was used to assessthe subjectrsquos verbal ability Performance was scored ac-cording to standardized procedures based on the sub-jectrsquos degree of familiarity with a list of up to 35 wordsand their elegance of expression The second test wasthe Block Design subtest comprising part of theWAIS-R Performance subtests which was used to assessthe subjectsrsquo constructional-spatial ability and was ad-ministered in the standardized manner

610 Analysis

A one-way repeated measures analysis of variance(ANOVA) was performed for each of the dependentvariables and the overall times and distances The with-in-subjects variable was the landmark description typewith three levels text image and worldlet Post hocanalyses were completed using the Scheffe test for dif-ferences in sample means Correlation analysis was per-

formed across the scores of the neuropsychological testsand the scores of the waynding tests We adopted asignicance level of 005 on all analyses unless otherwisenoted

7 Results

Our primary result was that people spend moretime consulting the worldlet guidebook than the otherguidebooks but that they require less time to reach thegoal when using the worldlet guidebook The followingspecic ndings tell us more about the component ac-tivities that benetted from worldlet use

71 Time and Distance Measures

Figures 8 through 11 show time and distancemeasures in which the independent variable (type ofguidebook) is displayed as white gray and black barsalong with the standard error of the mean (SEM) Fig-ure 8 shows the mean overall time and distance requiredto reach the goal kiosk and gure 9 breaks the overalltime into its components Figure 10 gives the forwardand backtracking distance components Figure 11 showsthe mean number of times that each type of guidebook

Figure 8 (left) Mean total time The overall time spent traveling to

the goal was signicantly less using the worldlet guidebook than when

using the text (F(1 24) 5 989 p 001) or image (F(1 24) 5

463 p 001) guidebooks (right) Mean total distance The overall

distance traveled using the worldlet guidebook was signicantly less

than the distance traveled using the text (F(1 24) 5 1141 p

001) or image (F(1 24) 5 893 p 001) guidebooks

576 PRESENCE VOLUME 10 NUMBER 6

was opened Figure 12 diagrams the subjectsrsquo scores onthe Wechsler neuropsychological tests

Analysis of variance followed by Scheffe post hoc testswere performed for the different measures Table 2summarizes these results

72 Street Plots

Figure 13 shows plots of the streets traversed by the26 subjects in the three test cities The thickness andbrightness of each street line indicate the total number oftimes that the street was traversed For example widerbrighter streets indicate heavily used avenues Backtrackingcan increase the number of street traversals beyond theexpected minimum of 26 The starting point and goalkiosk are indicated on each plot and landmarks enroute tothe goal are marked with a star (q )

Figure 10 The mean distance moving forward while using the

worldlet guidebook is signicantly less than the forward distance

covered while using the text or image guidebook Backtracking

distance is almost nonexistent for the worldlet guidebook task

Figure 9 Turning looking for landmark is time spent standing in

one location Turning changing forward direction is time spent

changing forward direction Consulting guidebook is time spent with

the guidebook on screen Moving forward is time spent traveling over

new ground and Backtracking is time spent retreating Note that the

time spent consulting the worldlet guidebook is signicantly greater

than the time spent consulting the text or image guidebook Also note

that the time spent backtracking is near zero when using the worldlet

guidebook

Figure 11 Subjects opened the text and image guidebooks an

average of fteen to sixteen times while seeking the eight landmarks

enroute to the goal kiosk The worldlet guidebook was opened eleven

times on average

Figure 12 Z-transformed spatial and verbal scores from the

neuropsychological tests The mean raw scores were 45 and 55 for

the spatial and verbal tests respectively

Elvins et al 577

73 Subjective Measures

As part of the written questionnaire administeredafter the experiment subjects were asked how difcultthe search task was using each of the different types ofguidebooks Although responses were scattered a cleartrend in favor of the worldlet guidebook was evident asshown in table 3

74 Correlation Analysis

One surprising nding was a high inverse corre-lation ( 2 077 p 5 001) between spatial skills andthe time spent consulting the text guidebook Sub-jects with better spatial ability seemed able to form amental picture of a landmark based on textual de-scription more quickly than did subjects with poorer

spatial ability This is not the case for image andworldlet guidebooks however

We expected to nd a negative correlation betweenspatial skills and overall travel time in the image andworldlet guidebook cities The data did not supportthis nor did they support another of our hypothesesthat there would be a positive correlation between ver-bal skills and the performance in the text guidebookcity Verbal skills made no difference in overall waynd-ing time

75 Observations on Subject Strategies

Many subjects said that the tasks were harderthan they had imagined and afterwards several saidthat they felt tired No subjects complained of mo-

Table 2 Our ANOVA Indicated Signicant Differences between All the Sample Means and Post Hoc Analysis Was Warranted

ANOVA

Post hoc comparisons (Scheffe F-test)

text vs image text vs worldlet image vs worldlet

TimeConsulting guidebook 5559 p 00001 1966 p 0001 884 p 001 5488 p 0011Looking for landmark 2161 p 00001 047 1869 p 0001 1325 p 001Changing heading 1736 p 00001 002 1353 p 001 1249 p 001Moving forward 1602 p 00001 0003 1220 p 001 1183 p 001Backtracking 955 p 00003 022 826 p 001 586 p 001Overall 1034 p 00002 098 989 p 001 463 p 001

DistanceMoving forward 1533 p 00001 005 1223 p 001 1072 p 001Backtracking 1019 p 00002 023 884 p 001 621 p 001Overall 1365 p 00001 015 1141 p 001 893 p 001

GuidebookOpened 1527 p 00001 104 759 p 001 1426 p 0001

Post hoc (Scheffe) analysis indicated for all dependent variables signicant differences in the sample means of the textversus worldlet performance and image versus worldlet performance Performance using the text versus imageguidebooks was not signicantly different except that the time spent consulting the text guidebook was signicantlygreater than the time spent consulting the image guidebook In each table cell the rst number is the F value and thesecond number is the p value Asterisk indicates signicance p 005 Time and distance measures are explained in thecaptions for gures 8 through 11

578 PRESENCE VOLUME 10 NUMBER 6

tion sickness Most subjects required 70 min or moreto nish all aspects of the computer-based part of theexperiment

We observed subjects to use three general types ofstrategies while deciding what direction to go from astreet intersection

c Look down each street from many viewpointswithin the intersection until a landmark or land-mark context is seen

c Look down each street and if nothing looks famil-iar search exhaustively in each direction until thenext landmark is found

c Look down each street and if nothing looks famil-iar try going one-half block in each direction andlook again

Subjects using the rst strategy seemed to nd theirway more quickly and with less backtracking than didsubjects who used the second and third strategies An-other strategy described by some subjects on their writ-ten questionnaire involved viewing a landmark in theworldlet guidebook from many distant viewpoints in thehopes that one of the viewpoints would be similar to thedirection from which they would actually approach the

Figure 13 In these street plots line thickness and brightness

indicates the number of times that a street was traversed In top and

middle the 26 subjects traversed some streets 47 times indicating

backtracking Top and middle also exhibit circuitous trails In bottom

very little backtracking occurred and circuitous trails are minimal

Table 3 Subjects Were Asked to Rate the Difculty of theSearch Task Using the Three Different Types of Guidebooks

Text Image Worldlet

Very easy 0 2 11Easy 10 8 9Doable 10 8 6Difcult 5 7 0Very difcult 1 1 0

Median Doable Doable Easy

Elvins et al 579

landmark in the city We observed that some subjectsoften forgot what it was they were looking for and sohad to open the guidebook momentarily to remindthemselves

Not surprisingly landmarks with prominent signage(such as ldquoBobrsquos Groceryrdquo) and landmarks painted withmore highly saturated colors seemed easier to nd thanother types of landmarks Also after passing all thelandmarks several subjects exclaimed that they had lostall sense of direction within the world

A nal unexpected result were the many similarities be-tween the text and image guidebooks The text and imageguidebooks were only signicantly different in the timesubjects spent consulting them Considering that text andimages represent landmarks in very different ways we ex-pected differences in time and distance measures

76 Major Findings

In interpreting these ndings it is important toremember that the subjects performed the wayndingtasks using only the landmark knowledge gained bystudying the guidebooks No procedural or survey(map) information was provided

c As we predicted the total time spent traveling tothe goal was signicantly less using the worldletguidebook than when using the text or imageguidebooks

c Subjects spent signicantly more time studying theworldlet guidebook than they did the text or imageguidebook

c Increased worldlet study time was more than com-pensated for by a reduction in the time required toreach the goal Using the worldlet guidebook sub-jects traveled an average of 55 faster than whenusing the text guidebook and 38 faster than whenusing the image guidebook

c Increased worldlet study time also allowed subjectsto move more directly to the goal and reducedbacktracking time to almost zero This is showndiagramatically in gure 13 in which gures 13aand 13b show brighter wider streets than in gure

13c Figure 13c also exhibits fewer detours and lesswandering than do gures 13a and 13b

We postulate that compared to the landmark knowl-edge gained from text and image guidebooks the im-proved landmark knowledge built while consulting theworldlet guidebook enabled subjects to refer less fre-quently to the guidebook This postulation is furthersupported by examining worldlet guidebook results

c Subjects spent signicantly more time studying theworldlet guidebook than they did the other types ofguidebooks

c When using the worldlet guidebook subjects trav-eled faster more directly and with more skill tothe goal kiosk

c Subjectsrsquo median rating for the difculty of thesearch task was Easy when using the worldlet guide-book

8 Conclusion

The waynding literature shows that landmarksare critical in navigating an environment whether it isreal or virtual While building landmark knowledgetravelers characterize a landmark representation by its3-D shape size texture context and so forth In awaynding task textual or image guidebook landmarkdescriptions fail to engage the full range of 3-D land-mark characteristics recognized and used by travelers tond their way

We know of no studies that have considered the com-parative advantages of different types of guidebooks forwaynding Intuitively one would predict that 3-Dthumbnails make better guidebooks for 3-D worldsthan do text and 2-D thumbnail images and indeedthis is what we found We wanted however to identifythe component activities in using guidebooks that mostbenetted from 3-D elements Perhaps of greatest inter-est was our nding that worldlets enabled subjects tond their way in complex environments with almost nobacktracking By spending more time studying theworldlets in their guidebooks they spent less time look-ing around reorienting themselves conducting exhaus-

580 PRESENCE VOLUME 10 NUMBER 6

tive searches and performing similar time-wasting activ-ities We must be cautious however in what weconclude because worldlets contain more informationthan do text or 2-D thumbnail images To control forthis difference in information content more experi-ments are necessary

9 Future Work

We would like to investigate the comparative ad-vantages of worldlet guidebooks for waynding tasks incorresponding virtual and real world environments(both urban and rural) We are also considering an ex-periment to evaluate tandem use of traditional 2-Dmaps and landmark guidebooks Requiring effectivemap and compass use in a natural environment thesport of orienteering (Kjellstrom 1976) might benetfrom worldlets By encapsulating the region around theagged control points into a worldlet orienteers couldgain rst-person experience with natural landmarks suchas foliage rocks gullies shadows and so on beforebeginning the search

Acknowledgments

Our appreciation goes to Planet 9 Studios (wwwplanet9com)for the use of their city building textures Thanks to JohnMoreland and Mike Bailey for technical assistance on thisproject and thanks to Suzi Feeney for collaborating in thepilot study design UCSD librarian Debbie Cox and her staffwere invaluable in locating references Thanks also to JohnHelly and Reagan Moore for their support San Diego Super-computer Center is funded by NSF the University of Califor-nia the State of California and industrial partners

References

Appleyard D A (1976) Planning a Pluralistic City Con-icting Realities in Ciudad Guayana Cambridge MAMIT Press

Art1 Com (1997) The cybercity Berlin project [on-line]Available httpwwwartcomdeprojects

Chen S E (1995) QuickTime VRmdashAn image-based ap-proach to VE navigation Proceedings of SIGGRAPH 9529 ndash38

Darken R P amp Sibert J L (1996) Waynding strategiesand behaviors in large virtual worlds Proceedings of CHI 96142ndash149

Downs R J amp Stea D (1973) Cognitive maps and spatialbehavior In R J Downs amp D Stea (Eds) Image and En-vironment Chicago Aldine Publishing Company 8 ndash26

Elvins T T (1997) Waynding in layered informationworlds Proceedings of the CODATA Euro-American Work-shop on Data and Information Visualization 6 ndash7

Elvins T T Nadeau D R amp Kirsh D (1997) World-letsmdash3D thumbnails for waynding in virtual environmentsProceedings of the ACM User Interface Software and Technol-ogy Symposium 21ndash30

Elvins T T (1998) WorldletsmdashA novel 3D thumbnail af-fordance for waynding in large virtual environmentsUnpublished doctoral dissertation Department of Elec-trical and Computer Engineering University of Califor-niandashSan Diego

Elvins T T Nadeau D R Kirsh D amp Schul R (1998)Worldletsmdash3D thumbnails for 3D browsing Proceedings ofthe ACM CHI Conference 21ndash30

Foley J van Dam A Feiner S amp Hughes J (1990) Com-puter Graphics Principles and Practice Reading MA Addi-son-Wesley

Gillner S amp Mallot H A (1998) Navigation and acquisi-tion of spatial knowledge in a virtual maze Journal of Cog-nitive Neuroscience 10(4) 445ndash 463

Goldin S E amp Thorndyke P W (1982) Simulating naviga-tion for spatial knowledge acquisition Human Factors24(4) 457ndash 471

ISO (1997) The Virtual Reality Modeling Language Inter-national Standard ISOIEC 14772-11997 [on-line] Avail-able httpwwwvrmlorgSpecicationsVRML97

Kjellstrom B (1976) Be Expert with Map and Compass TheComplete Orienteering Handbook New York Scribner Pub-lishers

Liggett R Friedman S amp Jepson W (1995) Interactivedesigndecision making in a virtual urban world visual sim-ulation and GIS Proceedings of the 1995 ESRI User Confer-ence Environmental Systems Research [on-line] Availablehttpwwwesricombasecommonuserconfproc95

Lynch K (1960) The Image of the City Cambridge MAMIT Press

Elvins et al 581

Thorndyke 1982) For a structure to be a landmark itmust have high imagability that is it must be distinctiveand memorable (Lynch 1960)

Landmarks are the subject of landmark knowledgebut they also play a part in survey and proceduralknowledge In survey knowledge landmarks provideregional anchors with which to calibrate distancesand directions In procedural knowledge landmarksmark decision points along a route helping in therecall of procedures to get to and from destinationsof interest

Whereas a travelerrsquos landmark knowledge charac-terizes a destination by its 3-D shape size textureand so forth traditional virtual environment (VE)browser menus and landmark guidebooks character-ize destinations only by textual descriptions or im-ages This representation mismatch reduces the effec-tiveness of destination menus and guidebooksUnable to engage their memory of 3-D landmarksto recognize destinations of interest travelers mayresort to a naive exhaustive search to nd a desiredlandmark

Figure 3 Construction of a worldlet A gas station landmark in a virtual city (top left) viewed from a

viewpoint in the street (top right) Viewed from above the gas station is upper left and a region

around the previous viewpoint is highlighted (bottom left) The gas station is captured within a spherical

worldlet and (bottom right) is viewed from street level within the worldlet

Elvins et al 567

21 Related Work

Although social and engineering scientists havestudied human waynding in the real world for manyyears (Appleyard 1976 Downs amp Stea 1973 Goldinamp Thorndyke 1982) only recently have computer sci-entists investigated this subject Satalich (1995) led sub-jects on a tour of a virtual building and then tested thesubjectsrsquo ability to nd their way to places seen duringthe tour Darken and Sibert (1996) applied Lynchrsquos(1960) city structure theories to large sparsely featuredVEs and then compared the subjectsrsquo ability to ndtheir way to target destinations with and without theadded city structure Witmer Bailey Knerr and Parsons(1996) studied route learning and found that subjectswho rehearsed a route in a virtual building learned theroute better than did subjects who rehearsed the routeverbally Gillner and Mallot (1997) found that subjectspieced together survey knowledge when repeatedly re-quired to nd shortest paths between a dozen unfamil-iar locations in a VE In studying VEs for real-worldtask training Waller Hunt and Knapp (1998) foundthat long exposure fostered good spatial knowledge un-der several VE delity conditions Ruddle Payne andJones (1998) studied path complexity and a compasscondition as factors in VE route learning and foundthat beyond two turns subjects were quickly disori-ented with or without a compass Although each ofthese studies recognized the value of landmark knowl-edge for waynding none studied the value of 3-Dlandmarks in familiarizing subjects with an environmentbefore entering it This is a primary goal of our study

22 Way nding Affordances

Three-dimensional models of large cities are be-coming available for exploration via the Web(Art 1 Com 1997 Liggett Friedman and Jepson1995 University College London 1997) As discussedpreviously travelers visiting such large VEs for the rsttime are easily disoriented may have difculty identify-ing a place upon arrival and may not be able to ndtheir way back to a place just visited (Downs amp Stea1973 Darken amp Sibert 1996) The following subsec-

tions describe affordances that travelers currently use toscrutinize potential landmark and VE destinations

c Textual descriptions are the dominant method usedto represent VE landmarks in world- and viewpoint-selection user interfaces HTML pages for instanceoften provide lists of available Web-based worlds(such as those authored in the Virtual Reality Mod-eling Language (VRML) (ISO 1997)) each onecharacterized by a URL a VE name andor a briefdescription Within VRML worlds textual descrip-tions characterize viewpoints and describe destina-tions associated with clickable anchor shapes

c Icons thumbnail images and image maps providecommon visual waynding aids In a 3-D contextgames often provide ldquojump gatesrdquo onto which im-ages of remote destinations are texture mappedStepping through such a gate jumps the traveler tothe destination landmark depicted on the gate

c Image mosaics group together multiple capturedimages into a traversable structure Applersquos Quick-Time VR for instance can use images capturedfrom multiple viewing angles at the same viewingposition (Chen 1995) By ordering images within atraveler-centered cylindrical structure QuickTimeVR can provide a traveler the ability to look in anydirection through automatic selection of an appro-priate image from the structure By chaining multi-ple mosaic structures together the content authorcan create a walkthrough path that hops from view-point to viewpoint Similar image mosaics can beused to create zoom paths pan paths and so forth

c Miniature Worlds and Maps are available in most3-D environment browsers and computer gamesand they enable the traveler to zoom out and viewthe world in miniature and to thereby gain surveyknowledge Pausch Burnette Brockway andWeiblen (1995) extend this notion by creating aworld in miniature (WIM) embedded within themain world The miniature world duplicates all ele-ments of the main world and adds an icon denotingthe travelerrsquos position and orientation Held withinthe travelerrsquos virtual hand the traveler can reachinto the miniature and reposition world content

568 PRESENCE VOLUME 10 NUMBER 6

andor themselves Simultaneously the outer mainworld is updated to match the altered miniatureautomatically adjusting the positions of shapes orthe traveler Similarly 2-D and 3-D maps are fre-quently found as navigation aids within VEs Forinstance 3-D games often provide a 3-D reduced-detail map in which an icon denotes the playerrsquoslocation Such maps can be panned zoomed androtated to provide alternate viewpoints similar tothat which is possible with miniature worlds

None of these waynding affordances offer the trav-eler faithful 3-D landmark representations or a rst-person experience with potential destinations Wickensand Baker (1995) argue that VE search tasks are dif-cult because ldquooften the object of the search is speciedin a format different from and more abstract than itsrendering in the VErdquo We address this problem by rep-resenting the object of the search in the same 3-D for-mat as the VE

3 What are Worldlets

In previous papers we described our ongoingworldlet waynding research (Elvins 1997 ElvinsNadeau amp Kirsh 1997 Elvins 1998 Elvins NadeauKirsh amp Schul 1998) Similar to a 3-D photograph aworldlet is associated with a viewing position and orien-tation and captures a snapshot of the 3-D shapes (usu-ally landmarks) that fall within the viewpointrsquos viewingvolume Worldlets can be explored and manipulated inthe same way that virtual worlds can be explored andmanipulated Incorporated into a VRML browser aguidebook or list of worldlets enables a world traveler tointeractively view landmarks and landmark context frommultiple viewpoints Seeing the landmarks presented inworldlets helps a traveler select destinations of interestand nd their way in unfamiliar worlds

In typical use the worldletrsquos viewpoint is aimed at animportant landmark and the worldletrsquos captured shapesreconstruct that landmark and its associated contextWhen viewed within an interactive 3-D browser a

worldlet provides a manipulable 3-D thumbnail repre-sentation of the landmark

Clicking on a worldlet in a landmark menu mightjump the traveler to the landmark represented by theworldlet The landmark could be at a viewpoint in thecurrent world or in another world A menu of worldletssuch as that shown in gure 4 might resemble land-mark history lists bookmarks or travel guidebooksTravelers are free to study the landmark in the worldletfrom any viewpoint before deciding whether to gothere

We have developed two types of worldlets

c A frustum worldlet contains shapes within a stan-dard pie-shaped viewing frustum positioned andoriented based upon a selected viewpoint Whenviewed a frustum worldlet looks like a pie-shapedfragment clipped from the larger main world

c A spherical worldlet contains shapes within a spheri-cal viewing bubble positioned at a selected view-point with a 360 deg eld of view When dis-played a spherical worldlet looks like a ball-shapedworld fragment similar to a snowglobe knick knackdepicting a favorite attraction

Figure 4 The navigation browserrsquos viewpoint selection window and

the waynding browserrsquos worldlet guidebook The ldquoGo tordquo button was

removed in the waynding browser

Elvins et al 569

For both worldlet types hither and yon clipping planesrestrict the extent of the worldlet ensuring that theworldlet contains a manageable subset of the largermain world Worldlet shape content is shaded and tex-tured to match the corresponding shapes in the mainworld Although the main world may have content thatchanges over time the captured worldlet remains staticrecording the content of the world at the time theworldlet was captured

Figure 3 depicts the process of constructing a spheri-cal worldlet from a fragment of a virtual city surround-ing a central viewpoint position Once captured world-lets can be browsed manipulated and explored in anumber of ways

4 Worldlets in Browsers

We have incorporated worldlets into the user in-terface for two experimental VRML browsers a naviga-tion browser and a waynding browser The navigationbrowser provides a ldquoGo tordquo feature that enables travel-ers to study select and y to world viewpoints in thecurrent world and to study select and load previouslyvisited worlds on the browserrsquos history list Guidebooksof worldlets in the waynding browser allow travelers tostudy 3-D landmark guidebooks before and duringtravel in an unfamiliar VE No go-to button is presentin this browser in which travelers manually move them-selves about based on the landmark knowledge gleanedduring study

41 Navigation Browser

Traditional VRML browsers provide a viewpointmenu that offers a choice of viewpoints each denotedby a brief textual description We have extended thisstandard feature to provide three experimental view-point selection interfaces each using worldlets All threepresent a set of worldlets one for each author-selectedviewpoint in the world A ldquoGo tordquo button ies the mainwindowrsquos viewpoint to that of the currently selectedworldlet The browser also supports on-the-y captureof worldlets from the travelerrsquos current viewpoint

c The viewpoint list window provides a list of world-lets beside a worldlet viewer Selection of a worldletfrom the list displays the worldlet in the viewerwhere it may be interactively panned zoomed androtated

c The viewpoint guidebook window presents a grid ofworldlet viewers arranged to form a guidebookphoto-album page Buttons on the window advancethe guidebook forward or back a page at a timeEach worldlet on the page can be interactively ex-amined Figure 4 shows the viewpoint guidebookwindow

c The viewpoint overlay window enables the travelerto select a worldlet from a list and overlay it atopthe main window highlighted in green This world-let overlay provides a clear indication of the world-letrsquos viewpoint position and orientation along withthe portion of the world captured within thatworldlet Figure 3b was generated using this overlaytechnique

Traditional VRML browsers provide a history list ofrecently visited worlds each denoted by its title orURL We have extended this standard feature to pro-vide two world selection interfaces each using world-lets In each world selection window a ldquoGo tordquo buttonloads into the main window the world associated withthe currently selected worldlet

c The world list window provides a list of worldletsbeside an interactive worldlet viewer similar to theviewpoint list window discussed earlier One world-let is available for each world on the browserrsquos his-tory list

c The world guidebook window uses the same guide-book photo-album layout used for the viewpointguidebook window discussed earlier One worldletis available for each world on the history list

A ldquoSave asrdquo feature of the VRML browser enables thetraveler to save a worldlet to a VRML le Using a col-lection of saved worldlets a world author can create aVRML world of worldlets Such a world acts like a 3-Ddestination index similar to a shelf full of snowglobeknick knacks When cast as a VRML anchor shape a

570 PRESENCE VOLUME 10 NUMBER 6

worldlet provides a 3-D button that when clicked onloads the associated world into the travelerrsquos browser

Figure 5 shows a world of worldlets Figure 5a shows

a close-up view of a world ldquodoorwayrdquo and a niche thatcontains a worldlet illustrating a viewpoint in thatworld Figure 5b shows a wider view of the same worldand multiple such doorways

42 Way nding Browser

The second browser was implemented specicallyfor the psychological experiment After manipulatingworldlets in a landmark guidebook to familiarize them-selves with the 3-D size shape color texture and con-text of the landmarks subjects in the experiment foundtheir way manually in the main window based upontheir landmark knowledge The landmark guidebookresembled the viewpoint selection window shown ingure 4 but it did not offer a ldquoGo tordquo button Thewaynding browser is described in detail in the methodssection

5 Implementation

Worldlets can be implemented in several waysThe implementation presented here is based on ahardware-accelerated algorithm and support algorithmsdeveloped specically for fast execution The VRMLbrowser used in this work maintains VE geometrywithin a tree-like scene graph Worldlets are also storedas scene graphs together with additional state informa-tion To capture a worldlet or to display a worldlet orVE the VRML browser traverses the associated scenegraph and feeds a 3-D graphics pipeline

51 Worldlet Capture in General

Any 3-D graphics pipeline can be roughly dividedinto two stages (i) transform clip and cull and (ii)rasterize (Foley van Dam Feiner amp Hughes 1990)The rst stage applies modeling viewing perspectiveand viewport transforms to map 3-D shapes to the 2-Dviewport Along the way shapes outside of the viewingfrustum are clipped away and backfaces are removedThe second stage uses 2-D shapes output by the rst

Figure 5 A world of worldlets that (top) associates a worldlet with

each doorway (bottom) in an environment containing multiple such

doorways Each doorway leads to a different world

Elvins et al 571

stage and draws the associated points lines and poly-gons on the screen

Worldlet capture taps into this 3-D graphics pipelineextracting the transformed shaded clipped and culledshape coordinates output by the rst stage prior to ras-terization in the second stage An extracted coordinatecontains x and y screen-space components a depth-buffer z-space component and the w coordinate Eachextracted coordinate has an associated RGB color andtexture coordinates which are computed by shadingand texture calculation phases in the rst pipeline stage

To create a worldlet these extracted coordinates areuntransformed to map them from viewport space backto world space as shown in gure 6 The inverses of the

viewport perspective viewing and modeling transformsare each applied Coordinate RGB colors and texturecoordinates are used to reconstruct 3-D worldlet geom-etry in a worldlet scene graph

Display of a worldlet passes this 3-D geometry backdown the graphics pipeline transforming clipping cull-ing and rasterizing the worldlet like any other 3-D con-tent

52 Frustum and Spherical Worldlets

A frustum worldlet is the result of capturing 3-Dgraphics pipeline output for a single traversal of thescene graph as viewed from the travelerrsquos current view-point The shape set extracted after the rst pipelinestage contains only those points lines and polygonsthat fall within the viewing frustum The worldlet con-structed by the browser from this geometry looks like apie-shaped slice cut out of the world

A spherical worldlet is the result of performing multi-ple frustum captures and combining the results TheVRML browser captures a spherical worldlet by spin-ning about a viewpoint position generating a set offrustum worldlets that each use a different viewing ori-entation Additional captures aimed straight up andstraight down complete the spherical worldlet The re-sulting set of capture geometry constructs a 360 degspherical view from the current viewpoint

When displayed the spherical worldletrsquos geometrylooks like a bubble cut out of the VE A close yon clipplane keeps the bubble small ensuring that it capturesonly landmark features in the immediate neighborhoodand not the entire virtual world

53 Worldlet Capture in OpenGL

To take advantage of the rendering speed offeredby the accelerated 3-D graphics pipeline on high-speedworkstations we implemented worldlet display and cap-ture using OpenInventor (Wernecke 1994) andOpenGL graphics libraries from Silicon Graphics Scenegraph construction and display traversal is managed byOpenInventor To capture worldlet geometry theVRML browser places the pipeline into feedback mode

Figure 6 World shapes are transformed to screen space via the

downward OpenGL graphics pipeline on the left Worldlet shapes are

inverse transformed from window coordinates to object coordinates via

the upward pipeline in the middle of the gure The intermediary

coordinate systems are named on the right

572 PRESENCE VOLUME 10 NUMBER 6

prior to a capture traversal and returns it to renderingmode following traversal

While in feedback mode the OpenGL pipeline divertsall transformed clipped and culled coordinates into abuffer provided by the browser Upon completion of acapture traversal no rasterization has taken place andthe feedback buffer contains the extracted geometry Byparsing the feedback buffer the VRML browser recon-structs worldlet geometry applying appropriate inversetransforms

OpenGL feedback buffer information includes shapecoordinates colors and texture coordinates but it doesnot include an indication of which texture image to usefor which bit of geometry To capture this additionalinformation the VRML browser uses OpenGLrsquos pass-through features to pass custom ags down through thepipeline during traversal To prepare these pass-through ags the browser augments the world scenegraph prior to traversal assigning a unique identier toeach texture image During a capture traversal eachtime a texture image is encountered the associatedidentier is passed down through the pipeline and intothe feedback buffer along with shape coordinates col-ors and texture coordinates During parsing of thefeedback buffer these texture identiers enable worldletgeometry reconstruction to apply the correct textureimages to the correct shapes

6 Methods

The psychological experiment was designed to testwhether guidebooks of worldlets are better for waynd-ing than guidebooks of text and images and if so why

61 General Description

A standard within-subject randomized designwas used Training began by reading each subject a se-ries of instructions and teaching the subject in a prac-tice virtual city (three blocks by three blocks) how tooperate the user interface Subjects were allowed tospend unlimited time in this city becoming familiar withthe controls and the online guidebooks

Each subject was then asked to nd his or her wayvia one landmark to the goal (a brightly colored kiosk)in three practice cities each ve blocks by ve blocks Inone city the guidebook represented each landmark witha paragraph of text in another city the landmarks wererepresented as 2-D thumbnail images and in a third citythe landmarks were represented as spherical worldletsThe guidebooks contained only landmark informationand subjects were given no procedural or survey infor-mation Although the image and worldlet guidebookhad the same appearance as shown in gure 4 theworldlets in the worldlet guidebook could be exploredand manipulated

After completing the practice tasks the experimentbegan and subjects were asked to nd their way to thegoal kiosk via eight ordered landmarks in three virtualcities ten blocks by ten blocks Performance in thesethree cities was timed and the subjectrsquos location orien-tation and actions were recorded at 1 sec intervalsSubjects visited the test cities in a random order tocounterbalance possible carryover effects and each ofthe cities and landmarks were different and unique

Before the computer portion of the experiment sub-jects completed a brief questionnaire on their prior ex-perience with computers After the experiment subjectsanswered written questions about the strategies theyused while nding their way and how difcult theyfound the task using each of the guidebooks Subjectsalso completed spatial and verbal neuropsychologicaltests during a separate session

62 Constants

The major independent variable in this experimentwas the type of guidebook used (text image or world-let) Parameters concerned with city design and land-mark layout as listed in table 1 were held constant

63 City Design

Our primary goal in designing the virtual citieswas to preserve as many elements of a real city as possi-ble Toward this goal we developed a pseudo-randomcity generator that produced cities based on a regular

Elvins et al 573

street grid with pavement roads and sidewalks betweenthe blocks Each block contained up to twenty build-ings side by side around the blockrsquos perimeter Using acache of 250 building designs buildings were randomlyselected and placed on city blocks Many non-landmarkbuildings were repeated between and within the citiesRendering requirements were reduced by programmingbuildings to be invisible when farther than three blocksfrom the current viewpoint To prevent subjects fromseeing buildings appearing and disappearing in the dis-tance we added fog to the environment

Cities contained parks parking lots and other civicfeatures but did not contain cars or street signs Land-marks were placed so that a subject could not short-cutthe route and would never have to make an exhaustivesearch as long as they carefully looked down each possi-ble path for familiar landmarks and landmark contextthey could always determine in which direction to pro-ceed No landmarks were located within one block ofthe perimeter wall that surrounded each city

64 Radius of Information

The radius of information was the most difcultparameter to make constant across the three guide-

books Given that text 2-D images and worldlets aredifferent representations they do not convey an equiva-lent quantity and quality of landmark information Wedetermined while building the guidebooks for the pilotstudy that using a radius of 20 m produced the mostnearly equivalent set of guidebooks For the text guide-book this meant describing in words the landmark thebuilding or structure directly across the street thebuilding on each side and whether the landmark was ona corner Thus our textual description of the landmarkshown in gure 3 read

Citgo Gas Station Citgorsquos red gas pumps are at acorner location Next door is the Cafe 360 in a beigecolored building and across the street is the LumbiniRestaurant in a red brick building

For the image guidebook a 20 m radius meant posi-tioning the camera across the street from the landmarkso that the captured thumbnail image would contain thebuildings or structures on the left and right of the land-mark For the worldlet guidebook setting the informa-tion radius simply meant setting the yon clip plane tocutaway 3-D shapes farther away than 20 m while cap-turing a spherical worldlet Image and worldlet camera

Table 1 Although the Arrangement of Buildings Landmarks and Routes Were Different in Each City Their Basic Structurein Terms of City Design and Landmark Layout was Kept Constant to Allow for Controlled Comparison Subjects Movedat a Fixed Speed in all Cities

Parameter Constant value Comments

City dimensions ten by ten blocks 100 total blocksBuilding lots per block twenty on perimeter of block building footprint variesDistance from start to goal 1200 meters past 100 12-meter-wide lotsLandmarks per city nine including goal kiosk ordered in guidebooksDistance between landmarks 15 blocks 6 025 blockTravel speed ve meterssecond slow driving speedTypes of landmarks eight one of each kind in each cityTurns required to reach goal ve left and right turnsRadius of information approximately twenty meters see city design sectionLandmark positions three on corner ve midblock goal is midblockGoal building yellowred striped kiosk six-sided viewpoint invariant

574 PRESENCE VOLUME 10 NUMBER 6

locations were selected so as not to suggest a directionof approach to the landmark

65 Hardware and Software Design

The experiment was conducted using a standard19 in monitor Stereo glasses were not used A maincity window displayed the city in the waynding VRMLbrowser Subjects used keyboard arrow keys to moveforward and backward a xed distance on each keypress and to turn left and right by a xed angle on eachkey press In the worldlet guidebook subjects used thearrow keys in the same manner to move around in theworldlet Collision detection prevented subjects frompassing through buildings and other objects in the city

Subjects pushed a Start button before beginning aLandmark button each time a landmark was reachedand a Stop button upon reaching the goal kiosk Sub-jects pushed a Guidebook button to display the guide-book Subjects practiced operating all of these buttonsin practice cities prior to beginning the experiment Thewindows were designed so that when the guidebookwas open (shown in gure 4) the main city window(shown in gure 7) was completely occluded and nomovement could be made

66 Lessons Learned from the PilotStudy

Our pilot study design and results are described byElvins et al (1997) From the pilot study we learnedthat landmarks cannot be more than two blocks apartbecause subjects frequently resort to an exhaustivesearch of the neighborhood We also learned that somesubjects will take much longer than other subjects tocomplete the task and that the task difculty and lengthmust be designed while keeping the slower wayndersin mind

67 Subject Instructions

The following is a subset of the instructions readto the subjects Instructions 3 and 4 were repeated be-fore beginning each city

1 The guidebook contains a series of landmarks thatyou probably want to follow since traveling pastthese landmarks will lead you to the goal kiosk

2 When you are at a landmark you may have towalk as far as one or two blocks before you will beable to see the next landmark

Figure 7 The main city window

Elvins et al 575

3 Your task is to locate the goal kiosk as quickly aspossible Go as fast as you can but be careful tonot get lost

4 Remember you can look at the guidebook as of-ten as you like

68 Subjects

Subjects were primarily University of CaliforniandashSan Diego students from a number of major depart-ments Ages ranged from 18 to 36 with a mean age of24 Fourteen men and twelve women completed theexperiment Results from four other subjects had to beexcluded from our data because the subjects becamelost in one or more of the cities All subjects were com-puter literate but with varying degrees of experiencewith VEs

69 Neuropsychological Tests

To provide an independent measure of a subset ofthe subjectsrsquo cognitive abilities subjects were tested in aseparate session using two standardized subtests fromthe Wechsler Adult Intelligent Scale-Revised (WAIS-R)(Wechsler 1955) The Vocabulary subtest comprisingpart of the WAIS-R Verbal subtests was used to assessthe subjectrsquos verbal ability Performance was scored ac-cording to standardized procedures based on the sub-jectrsquos degree of familiarity with a list of up to 35 wordsand their elegance of expression The second test wasthe Block Design subtest comprising part of theWAIS-R Performance subtests which was used to assessthe subjectsrsquo constructional-spatial ability and was ad-ministered in the standardized manner

610 Analysis

A one-way repeated measures analysis of variance(ANOVA) was performed for each of the dependentvariables and the overall times and distances The with-in-subjects variable was the landmark description typewith three levels text image and worldlet Post hocanalyses were completed using the Scheffe test for dif-ferences in sample means Correlation analysis was per-

formed across the scores of the neuropsychological testsand the scores of the waynding tests We adopted asignicance level of 005 on all analyses unless otherwisenoted

7 Results

Our primary result was that people spend moretime consulting the worldlet guidebook than the otherguidebooks but that they require less time to reach thegoal when using the worldlet guidebook The followingspecic ndings tell us more about the component ac-tivities that benetted from worldlet use

71 Time and Distance Measures

Figures 8 through 11 show time and distancemeasures in which the independent variable (type ofguidebook) is displayed as white gray and black barsalong with the standard error of the mean (SEM) Fig-ure 8 shows the mean overall time and distance requiredto reach the goal kiosk and gure 9 breaks the overalltime into its components Figure 10 gives the forwardand backtracking distance components Figure 11 showsthe mean number of times that each type of guidebook

Figure 8 (left) Mean total time The overall time spent traveling to

the goal was signicantly less using the worldlet guidebook than when

using the text (F(1 24) 5 989 p 001) or image (F(1 24) 5

463 p 001) guidebooks (right) Mean total distance The overall

distance traveled using the worldlet guidebook was signicantly less

than the distance traveled using the text (F(1 24) 5 1141 p

001) or image (F(1 24) 5 893 p 001) guidebooks

576 PRESENCE VOLUME 10 NUMBER 6

was opened Figure 12 diagrams the subjectsrsquo scores onthe Wechsler neuropsychological tests

Analysis of variance followed by Scheffe post hoc testswere performed for the different measures Table 2summarizes these results

72 Street Plots

Figure 13 shows plots of the streets traversed by the26 subjects in the three test cities The thickness andbrightness of each street line indicate the total number oftimes that the street was traversed For example widerbrighter streets indicate heavily used avenues Backtrackingcan increase the number of street traversals beyond theexpected minimum of 26 The starting point and goalkiosk are indicated on each plot and landmarks enroute tothe goal are marked with a star (q )

Figure 10 The mean distance moving forward while using the

worldlet guidebook is signicantly less than the forward distance

covered while using the text or image guidebook Backtracking

distance is almost nonexistent for the worldlet guidebook task

Figure 9 Turning looking for landmark is time spent standing in

one location Turning changing forward direction is time spent

changing forward direction Consulting guidebook is time spent with

the guidebook on screen Moving forward is time spent traveling over

new ground and Backtracking is time spent retreating Note that the

time spent consulting the worldlet guidebook is signicantly greater

than the time spent consulting the text or image guidebook Also note

that the time spent backtracking is near zero when using the worldlet

guidebook

Figure 11 Subjects opened the text and image guidebooks an

average of fteen to sixteen times while seeking the eight landmarks

enroute to the goal kiosk The worldlet guidebook was opened eleven

times on average

Figure 12 Z-transformed spatial and verbal scores from the

neuropsychological tests The mean raw scores were 45 and 55 for

the spatial and verbal tests respectively

Elvins et al 577

73 Subjective Measures

As part of the written questionnaire administeredafter the experiment subjects were asked how difcultthe search task was using each of the different types ofguidebooks Although responses were scattered a cleartrend in favor of the worldlet guidebook was evident asshown in table 3

74 Correlation Analysis

One surprising nding was a high inverse corre-lation ( 2 077 p 5 001) between spatial skills andthe time spent consulting the text guidebook Sub-jects with better spatial ability seemed able to form amental picture of a landmark based on textual de-scription more quickly than did subjects with poorer

spatial ability This is not the case for image andworldlet guidebooks however

We expected to nd a negative correlation betweenspatial skills and overall travel time in the image andworldlet guidebook cities The data did not supportthis nor did they support another of our hypothesesthat there would be a positive correlation between ver-bal skills and the performance in the text guidebookcity Verbal skills made no difference in overall waynd-ing time

75 Observations on Subject Strategies

Many subjects said that the tasks were harderthan they had imagined and afterwards several saidthat they felt tired No subjects complained of mo-

Table 2 Our ANOVA Indicated Signicant Differences between All the Sample Means and Post Hoc Analysis Was Warranted

ANOVA

Post hoc comparisons (Scheffe F-test)

text vs image text vs worldlet image vs worldlet

TimeConsulting guidebook 5559 p 00001 1966 p 0001 884 p 001 5488 p 0011Looking for landmark 2161 p 00001 047 1869 p 0001 1325 p 001Changing heading 1736 p 00001 002 1353 p 001 1249 p 001Moving forward 1602 p 00001 0003 1220 p 001 1183 p 001Backtracking 955 p 00003 022 826 p 001 586 p 001Overall 1034 p 00002 098 989 p 001 463 p 001

DistanceMoving forward 1533 p 00001 005 1223 p 001 1072 p 001Backtracking 1019 p 00002 023 884 p 001 621 p 001Overall 1365 p 00001 015 1141 p 001 893 p 001

GuidebookOpened 1527 p 00001 104 759 p 001 1426 p 0001

Post hoc (Scheffe) analysis indicated for all dependent variables signicant differences in the sample means of the textversus worldlet performance and image versus worldlet performance Performance using the text versus imageguidebooks was not signicantly different except that the time spent consulting the text guidebook was signicantlygreater than the time spent consulting the image guidebook In each table cell the rst number is the F value and thesecond number is the p value Asterisk indicates signicance p 005 Time and distance measures are explained in thecaptions for gures 8 through 11

578 PRESENCE VOLUME 10 NUMBER 6

tion sickness Most subjects required 70 min or moreto nish all aspects of the computer-based part of theexperiment

We observed subjects to use three general types ofstrategies while deciding what direction to go from astreet intersection

c Look down each street from many viewpointswithin the intersection until a landmark or land-mark context is seen

c Look down each street and if nothing looks famil-iar search exhaustively in each direction until thenext landmark is found

c Look down each street and if nothing looks famil-iar try going one-half block in each direction andlook again

Subjects using the rst strategy seemed to nd theirway more quickly and with less backtracking than didsubjects who used the second and third strategies An-other strategy described by some subjects on their writ-ten questionnaire involved viewing a landmark in theworldlet guidebook from many distant viewpoints in thehopes that one of the viewpoints would be similar to thedirection from which they would actually approach the

Figure 13 In these street plots line thickness and brightness

indicates the number of times that a street was traversed In top and

middle the 26 subjects traversed some streets 47 times indicating

backtracking Top and middle also exhibit circuitous trails In bottom

very little backtracking occurred and circuitous trails are minimal

Table 3 Subjects Were Asked to Rate the Difculty of theSearch Task Using the Three Different Types of Guidebooks

Text Image Worldlet

Very easy 0 2 11Easy 10 8 9Doable 10 8 6Difcult 5 7 0Very difcult 1 1 0

Median Doable Doable Easy

Elvins et al 579

landmark in the city We observed that some subjectsoften forgot what it was they were looking for and sohad to open the guidebook momentarily to remindthemselves

Not surprisingly landmarks with prominent signage(such as ldquoBobrsquos Groceryrdquo) and landmarks painted withmore highly saturated colors seemed easier to nd thanother types of landmarks Also after passing all thelandmarks several subjects exclaimed that they had lostall sense of direction within the world

A nal unexpected result were the many similarities be-tween the text and image guidebooks The text and imageguidebooks were only signicantly different in the timesubjects spent consulting them Considering that text andimages represent landmarks in very different ways we ex-pected differences in time and distance measures

76 Major Findings

In interpreting these ndings it is important toremember that the subjects performed the wayndingtasks using only the landmark knowledge gained bystudying the guidebooks No procedural or survey(map) information was provided

c As we predicted the total time spent traveling tothe goal was signicantly less using the worldletguidebook than when using the text or imageguidebooks

c Subjects spent signicantly more time studying theworldlet guidebook than they did the text or imageguidebook

c Increased worldlet study time was more than com-pensated for by a reduction in the time required toreach the goal Using the worldlet guidebook sub-jects traveled an average of 55 faster than whenusing the text guidebook and 38 faster than whenusing the image guidebook

c Increased worldlet study time also allowed subjectsto move more directly to the goal and reducedbacktracking time to almost zero This is showndiagramatically in gure 13 in which gures 13aand 13b show brighter wider streets than in gure

13c Figure 13c also exhibits fewer detours and lesswandering than do gures 13a and 13b

We postulate that compared to the landmark knowl-edge gained from text and image guidebooks the im-proved landmark knowledge built while consulting theworldlet guidebook enabled subjects to refer less fre-quently to the guidebook This postulation is furthersupported by examining worldlet guidebook results

c Subjects spent signicantly more time studying theworldlet guidebook than they did the other types ofguidebooks

c When using the worldlet guidebook subjects trav-eled faster more directly and with more skill tothe goal kiosk

c Subjectsrsquo median rating for the difculty of thesearch task was Easy when using the worldlet guide-book

8 Conclusion

The waynding literature shows that landmarksare critical in navigating an environment whether it isreal or virtual While building landmark knowledgetravelers characterize a landmark representation by its3-D shape size texture context and so forth In awaynding task textual or image guidebook landmarkdescriptions fail to engage the full range of 3-D land-mark characteristics recognized and used by travelers tond their way

We know of no studies that have considered the com-parative advantages of different types of guidebooks forwaynding Intuitively one would predict that 3-Dthumbnails make better guidebooks for 3-D worldsthan do text and 2-D thumbnail images and indeedthis is what we found We wanted however to identifythe component activities in using guidebooks that mostbenetted from 3-D elements Perhaps of greatest inter-est was our nding that worldlets enabled subjects tond their way in complex environments with almost nobacktracking By spending more time studying theworldlets in their guidebooks they spent less time look-ing around reorienting themselves conducting exhaus-

580 PRESENCE VOLUME 10 NUMBER 6

tive searches and performing similar time-wasting activ-ities We must be cautious however in what weconclude because worldlets contain more informationthan do text or 2-D thumbnail images To control forthis difference in information content more experi-ments are necessary

9 Future Work

We would like to investigate the comparative ad-vantages of worldlet guidebooks for waynding tasks incorresponding virtual and real world environments(both urban and rural) We are also considering an ex-periment to evaluate tandem use of traditional 2-Dmaps and landmark guidebooks Requiring effectivemap and compass use in a natural environment thesport of orienteering (Kjellstrom 1976) might benetfrom worldlets By encapsulating the region around theagged control points into a worldlet orienteers couldgain rst-person experience with natural landmarks suchas foliage rocks gullies shadows and so on beforebeginning the search

Acknowledgments

Our appreciation goes to Planet 9 Studios (wwwplanet9com)for the use of their city building textures Thanks to JohnMoreland and Mike Bailey for technical assistance on thisproject and thanks to Suzi Feeney for collaborating in thepilot study design UCSD librarian Debbie Cox and her staffwere invaluable in locating references Thanks also to JohnHelly and Reagan Moore for their support San Diego Super-computer Center is funded by NSF the University of Califor-nia the State of California and industrial partners

References

Appleyard D A (1976) Planning a Pluralistic City Con-icting Realities in Ciudad Guayana Cambridge MAMIT Press

Art1 Com (1997) The cybercity Berlin project [on-line]Available httpwwwartcomdeprojects

Chen S E (1995) QuickTime VRmdashAn image-based ap-proach to VE navigation Proceedings of SIGGRAPH 9529 ndash38

Darken R P amp Sibert J L (1996) Waynding strategiesand behaviors in large virtual worlds Proceedings of CHI 96142ndash149

Downs R J amp Stea D (1973) Cognitive maps and spatialbehavior In R J Downs amp D Stea (Eds) Image and En-vironment Chicago Aldine Publishing Company 8 ndash26

Elvins T T (1997) Waynding in layered informationworlds Proceedings of the CODATA Euro-American Work-shop on Data and Information Visualization 6 ndash7

Elvins T T Nadeau D R amp Kirsh D (1997) World-letsmdash3D thumbnails for waynding in virtual environmentsProceedings of the ACM User Interface Software and Technol-ogy Symposium 21ndash30

Elvins T T (1998) WorldletsmdashA novel 3D thumbnail af-fordance for waynding in large virtual environmentsUnpublished doctoral dissertation Department of Elec-trical and Computer Engineering University of Califor-niandashSan Diego

Elvins T T Nadeau D R Kirsh D amp Schul R (1998)Worldletsmdash3D thumbnails for 3D browsing Proceedings ofthe ACM CHI Conference 21ndash30

Foley J van Dam A Feiner S amp Hughes J (1990) Com-puter Graphics Principles and Practice Reading MA Addi-son-Wesley

Gillner S amp Mallot H A (1998) Navigation and acquisi-tion of spatial knowledge in a virtual maze Journal of Cog-nitive Neuroscience 10(4) 445ndash 463

Goldin S E amp Thorndyke P W (1982) Simulating naviga-tion for spatial knowledge acquisition Human Factors24(4) 457ndash 471

ISO (1997) The Virtual Reality Modeling Language Inter-national Standard ISOIEC 14772-11997 [on-line] Avail-able httpwwwvrmlorgSpecicationsVRML97

Kjellstrom B (1976) Be Expert with Map and Compass TheComplete Orienteering Handbook New York Scribner Pub-lishers

Liggett R Friedman S amp Jepson W (1995) Interactivedesigndecision making in a virtual urban world visual sim-ulation and GIS Proceedings of the 1995 ESRI User Confer-ence Environmental Systems Research [on-line] Availablehttpwwwesricombasecommonuserconfproc95

Lynch K (1960) The Image of the City Cambridge MAMIT Press

Elvins et al 581

21 Related Work

Although social and engineering scientists havestudied human waynding in the real world for manyyears (Appleyard 1976 Downs amp Stea 1973 Goldinamp Thorndyke 1982) only recently have computer sci-entists investigated this subject Satalich (1995) led sub-jects on a tour of a virtual building and then tested thesubjectsrsquo ability to nd their way to places seen duringthe tour Darken and Sibert (1996) applied Lynchrsquos(1960) city structure theories to large sparsely featuredVEs and then compared the subjectsrsquo ability to ndtheir way to target destinations with and without theadded city structure Witmer Bailey Knerr and Parsons(1996) studied route learning and found that subjectswho rehearsed a route in a virtual building learned theroute better than did subjects who rehearsed the routeverbally Gillner and Mallot (1997) found that subjectspieced together survey knowledge when repeatedly re-quired to nd shortest paths between a dozen unfamil-iar locations in a VE In studying VEs for real-worldtask training Waller Hunt and Knapp (1998) foundthat long exposure fostered good spatial knowledge un-der several VE delity conditions Ruddle Payne andJones (1998) studied path complexity and a compasscondition as factors in VE route learning and foundthat beyond two turns subjects were quickly disori-ented with or without a compass Although each ofthese studies recognized the value of landmark knowl-edge for waynding none studied the value of 3-Dlandmarks in familiarizing subjects with an environmentbefore entering it This is a primary goal of our study

22 Way nding Affordances

Three-dimensional models of large cities are be-coming available for exploration via the Web(Art 1 Com 1997 Liggett Friedman and Jepson1995 University College London 1997) As discussedpreviously travelers visiting such large VEs for the rsttime are easily disoriented may have difculty identify-ing a place upon arrival and may not be able to ndtheir way back to a place just visited (Downs amp Stea1973 Darken amp Sibert 1996) The following subsec-

tions describe affordances that travelers currently use toscrutinize potential landmark and VE destinations

c Textual descriptions are the dominant method usedto represent VE landmarks in world- and viewpoint-selection user interfaces HTML pages for instanceoften provide lists of available Web-based worlds(such as those authored in the Virtual Reality Mod-eling Language (VRML) (ISO 1997)) each onecharacterized by a URL a VE name andor a briefdescription Within VRML worlds textual descrip-tions characterize viewpoints and describe destina-tions associated with clickable anchor shapes

c Icons thumbnail images and image maps providecommon visual waynding aids In a 3-D contextgames often provide ldquojump gatesrdquo onto which im-ages of remote destinations are texture mappedStepping through such a gate jumps the traveler tothe destination landmark depicted on the gate

c Image mosaics group together multiple capturedimages into a traversable structure Applersquos Quick-Time VR for instance can use images capturedfrom multiple viewing angles at the same viewingposition (Chen 1995) By ordering images within atraveler-centered cylindrical structure QuickTimeVR can provide a traveler the ability to look in anydirection through automatic selection of an appro-priate image from the structure By chaining multi-ple mosaic structures together the content authorcan create a walkthrough path that hops from view-point to viewpoint Similar image mosaics can beused to create zoom paths pan paths and so forth

c Miniature Worlds and Maps are available in most3-D environment browsers and computer gamesand they enable the traveler to zoom out and viewthe world in miniature and to thereby gain surveyknowledge Pausch Burnette Brockway andWeiblen (1995) extend this notion by creating aworld in miniature (WIM) embedded within themain world The miniature world duplicates all ele-ments of the main world and adds an icon denotingthe travelerrsquos position and orientation Held withinthe travelerrsquos virtual hand the traveler can reachinto the miniature and reposition world content

568 PRESENCE VOLUME 10 NUMBER 6

andor themselves Simultaneously the outer mainworld is updated to match the altered miniatureautomatically adjusting the positions of shapes orthe traveler Similarly 2-D and 3-D maps are fre-quently found as navigation aids within VEs Forinstance 3-D games often provide a 3-D reduced-detail map in which an icon denotes the playerrsquoslocation Such maps can be panned zoomed androtated to provide alternate viewpoints similar tothat which is possible with miniature worlds

None of these waynding affordances offer the trav-eler faithful 3-D landmark representations or a rst-person experience with potential destinations Wickensand Baker (1995) argue that VE search tasks are dif-cult because ldquooften the object of the search is speciedin a format different from and more abstract than itsrendering in the VErdquo We address this problem by rep-resenting the object of the search in the same 3-D for-mat as the VE

3 What are Worldlets

In previous papers we described our ongoingworldlet waynding research (Elvins 1997 ElvinsNadeau amp Kirsh 1997 Elvins 1998 Elvins NadeauKirsh amp Schul 1998) Similar to a 3-D photograph aworldlet is associated with a viewing position and orien-tation and captures a snapshot of the 3-D shapes (usu-ally landmarks) that fall within the viewpointrsquos viewingvolume Worldlets can be explored and manipulated inthe same way that virtual worlds can be explored andmanipulated Incorporated into a VRML browser aguidebook or list of worldlets enables a world traveler tointeractively view landmarks and landmark context frommultiple viewpoints Seeing the landmarks presented inworldlets helps a traveler select destinations of interestand nd their way in unfamiliar worlds

In typical use the worldletrsquos viewpoint is aimed at animportant landmark and the worldletrsquos captured shapesreconstruct that landmark and its associated contextWhen viewed within an interactive 3-D browser a

worldlet provides a manipulable 3-D thumbnail repre-sentation of the landmark

Clicking on a worldlet in a landmark menu mightjump the traveler to the landmark represented by theworldlet The landmark could be at a viewpoint in thecurrent world or in another world A menu of worldletssuch as that shown in gure 4 might resemble land-mark history lists bookmarks or travel guidebooksTravelers are free to study the landmark in the worldletfrom any viewpoint before deciding whether to gothere

We have developed two types of worldlets

c A frustum worldlet contains shapes within a stan-dard pie-shaped viewing frustum positioned andoriented based upon a selected viewpoint Whenviewed a frustum worldlet looks like a pie-shapedfragment clipped from the larger main world

c A spherical worldlet contains shapes within a spheri-cal viewing bubble positioned at a selected view-point with a 360 deg eld of view When dis-played a spherical worldlet looks like a ball-shapedworld fragment similar to a snowglobe knick knackdepicting a favorite attraction

Figure 4 The navigation browserrsquos viewpoint selection window and

the waynding browserrsquos worldlet guidebook The ldquoGo tordquo button was

removed in the waynding browser

Elvins et al 569

For both worldlet types hither and yon clipping planesrestrict the extent of the worldlet ensuring that theworldlet contains a manageable subset of the largermain world Worldlet shape content is shaded and tex-tured to match the corresponding shapes in the mainworld Although the main world may have content thatchanges over time the captured worldlet remains staticrecording the content of the world at the time theworldlet was captured

Figure 3 depicts the process of constructing a spheri-cal worldlet from a fragment of a virtual city surround-ing a central viewpoint position Once captured world-lets can be browsed manipulated and explored in anumber of ways

4 Worldlets in Browsers

We have incorporated worldlets into the user in-terface for two experimental VRML browsers a naviga-tion browser and a waynding browser The navigationbrowser provides a ldquoGo tordquo feature that enables travel-ers to study select and y to world viewpoints in thecurrent world and to study select and load previouslyvisited worlds on the browserrsquos history list Guidebooksof worldlets in the waynding browser allow travelers tostudy 3-D landmark guidebooks before and duringtravel in an unfamiliar VE No go-to button is presentin this browser in which travelers manually move them-selves about based on the landmark knowledge gleanedduring study

41 Navigation Browser

Traditional VRML browsers provide a viewpointmenu that offers a choice of viewpoints each denotedby a brief textual description We have extended thisstandard feature to provide three experimental view-point selection interfaces each using worldlets All threepresent a set of worldlets one for each author-selectedviewpoint in the world A ldquoGo tordquo button ies the mainwindowrsquos viewpoint to that of the currently selectedworldlet The browser also supports on-the-y captureof worldlets from the travelerrsquos current viewpoint

c The viewpoint list window provides a list of world-lets beside a worldlet viewer Selection of a worldletfrom the list displays the worldlet in the viewerwhere it may be interactively panned zoomed androtated

c The viewpoint guidebook window presents a grid ofworldlet viewers arranged to form a guidebookphoto-album page Buttons on the window advancethe guidebook forward or back a page at a timeEach worldlet on the page can be interactively ex-amined Figure 4 shows the viewpoint guidebookwindow

c The viewpoint overlay window enables the travelerto select a worldlet from a list and overlay it atopthe main window highlighted in green This world-let overlay provides a clear indication of the world-letrsquos viewpoint position and orientation along withthe portion of the world captured within thatworldlet Figure 3b was generated using this overlaytechnique

Traditional VRML browsers provide a history list ofrecently visited worlds each denoted by its title orURL We have extended this standard feature to pro-vide two world selection interfaces each using world-lets In each world selection window a ldquoGo tordquo buttonloads into the main window the world associated withthe currently selected worldlet

c The world list window provides a list of worldletsbeside an interactive worldlet viewer similar to theviewpoint list window discussed earlier One world-let is available for each world on the browserrsquos his-tory list

c The world guidebook window uses the same guide-book photo-album layout used for the viewpointguidebook window discussed earlier One worldletis available for each world on the history list

A ldquoSave asrdquo feature of the VRML browser enables thetraveler to save a worldlet to a VRML le Using a col-lection of saved worldlets a world author can create aVRML world of worldlets Such a world acts like a 3-Ddestination index similar to a shelf full of snowglobeknick knacks When cast as a VRML anchor shape a

570 PRESENCE VOLUME 10 NUMBER 6

worldlet provides a 3-D button that when clicked onloads the associated world into the travelerrsquos browser

Figure 5 shows a world of worldlets Figure 5a shows

a close-up view of a world ldquodoorwayrdquo and a niche thatcontains a worldlet illustrating a viewpoint in thatworld Figure 5b shows a wider view of the same worldand multiple such doorways

42 Way nding Browser

The second browser was implemented specicallyfor the psychological experiment After manipulatingworldlets in a landmark guidebook to familiarize them-selves with the 3-D size shape color texture and con-text of the landmarks subjects in the experiment foundtheir way manually in the main window based upontheir landmark knowledge The landmark guidebookresembled the viewpoint selection window shown ingure 4 but it did not offer a ldquoGo tordquo button Thewaynding browser is described in detail in the methodssection

5 Implementation

Worldlets can be implemented in several waysThe implementation presented here is based on ahardware-accelerated algorithm and support algorithmsdeveloped specically for fast execution The VRMLbrowser used in this work maintains VE geometrywithin a tree-like scene graph Worldlets are also storedas scene graphs together with additional state informa-tion To capture a worldlet or to display a worldlet orVE the VRML browser traverses the associated scenegraph and feeds a 3-D graphics pipeline

51 Worldlet Capture in General

Any 3-D graphics pipeline can be roughly dividedinto two stages (i) transform clip and cull and (ii)rasterize (Foley van Dam Feiner amp Hughes 1990)The rst stage applies modeling viewing perspectiveand viewport transforms to map 3-D shapes to the 2-Dviewport Along the way shapes outside of the viewingfrustum are clipped away and backfaces are removedThe second stage uses 2-D shapes output by the rst

Figure 5 A world of worldlets that (top) associates a worldlet with

each doorway (bottom) in an environment containing multiple such

doorways Each doorway leads to a different world

Elvins et al 571

stage and draws the associated points lines and poly-gons on the screen

Worldlet capture taps into this 3-D graphics pipelineextracting the transformed shaded clipped and culledshape coordinates output by the rst stage prior to ras-terization in the second stage An extracted coordinatecontains x and y screen-space components a depth-buffer z-space component and the w coordinate Eachextracted coordinate has an associated RGB color andtexture coordinates which are computed by shadingand texture calculation phases in the rst pipeline stage

To create a worldlet these extracted coordinates areuntransformed to map them from viewport space backto world space as shown in gure 6 The inverses of the

viewport perspective viewing and modeling transformsare each applied Coordinate RGB colors and texturecoordinates are used to reconstruct 3-D worldlet geom-etry in a worldlet scene graph

Display of a worldlet passes this 3-D geometry backdown the graphics pipeline transforming clipping cull-ing and rasterizing the worldlet like any other 3-D con-tent

52 Frustum and Spherical Worldlets

A frustum worldlet is the result of capturing 3-Dgraphics pipeline output for a single traversal of thescene graph as viewed from the travelerrsquos current view-point The shape set extracted after the rst pipelinestage contains only those points lines and polygonsthat fall within the viewing frustum The worldlet con-structed by the browser from this geometry looks like apie-shaped slice cut out of the world

A spherical worldlet is the result of performing multi-ple frustum captures and combining the results TheVRML browser captures a spherical worldlet by spin-ning about a viewpoint position generating a set offrustum worldlets that each use a different viewing ori-entation Additional captures aimed straight up andstraight down complete the spherical worldlet The re-sulting set of capture geometry constructs a 360 degspherical view from the current viewpoint

When displayed the spherical worldletrsquos geometrylooks like a bubble cut out of the VE A close yon clipplane keeps the bubble small ensuring that it capturesonly landmark features in the immediate neighborhoodand not the entire virtual world

53 Worldlet Capture in OpenGL

To take advantage of the rendering speed offeredby the accelerated 3-D graphics pipeline on high-speedworkstations we implemented worldlet display and cap-ture using OpenInventor (Wernecke 1994) andOpenGL graphics libraries from Silicon Graphics Scenegraph construction and display traversal is managed byOpenInventor To capture worldlet geometry theVRML browser places the pipeline into feedback mode

Figure 6 World shapes are transformed to screen space via the

downward OpenGL graphics pipeline on the left Worldlet shapes are

inverse transformed from window coordinates to object coordinates via

the upward pipeline in the middle of the gure The intermediary

coordinate systems are named on the right

572 PRESENCE VOLUME 10 NUMBER 6

prior to a capture traversal and returns it to renderingmode following traversal

While in feedback mode the OpenGL pipeline divertsall transformed clipped and culled coordinates into abuffer provided by the browser Upon completion of acapture traversal no rasterization has taken place andthe feedback buffer contains the extracted geometry Byparsing the feedback buffer the VRML browser recon-structs worldlet geometry applying appropriate inversetransforms

OpenGL feedback buffer information includes shapecoordinates colors and texture coordinates but it doesnot include an indication of which texture image to usefor which bit of geometry To capture this additionalinformation the VRML browser uses OpenGLrsquos pass-through features to pass custom ags down through thepipeline during traversal To prepare these pass-through ags the browser augments the world scenegraph prior to traversal assigning a unique identier toeach texture image During a capture traversal eachtime a texture image is encountered the associatedidentier is passed down through the pipeline and intothe feedback buffer along with shape coordinates col-ors and texture coordinates During parsing of thefeedback buffer these texture identiers enable worldletgeometry reconstruction to apply the correct textureimages to the correct shapes

6 Methods

The psychological experiment was designed to testwhether guidebooks of worldlets are better for waynd-ing than guidebooks of text and images and if so why

61 General Description

A standard within-subject randomized designwas used Training began by reading each subject a se-ries of instructions and teaching the subject in a prac-tice virtual city (three blocks by three blocks) how tooperate the user interface Subjects were allowed tospend unlimited time in this city becoming familiar withthe controls and the online guidebooks

Each subject was then asked to nd his or her wayvia one landmark to the goal (a brightly colored kiosk)in three practice cities each ve blocks by ve blocks Inone city the guidebook represented each landmark witha paragraph of text in another city the landmarks wererepresented as 2-D thumbnail images and in a third citythe landmarks were represented as spherical worldletsThe guidebooks contained only landmark informationand subjects were given no procedural or survey infor-mation Although the image and worldlet guidebookhad the same appearance as shown in gure 4 theworldlets in the worldlet guidebook could be exploredand manipulated

After completing the practice tasks the experimentbegan and subjects were asked to nd their way to thegoal kiosk via eight ordered landmarks in three virtualcities ten blocks by ten blocks Performance in thesethree cities was timed and the subjectrsquos location orien-tation and actions were recorded at 1 sec intervalsSubjects visited the test cities in a random order tocounterbalance possible carryover effects and each ofthe cities and landmarks were different and unique

Before the computer portion of the experiment sub-jects completed a brief questionnaire on their prior ex-perience with computers After the experiment subjectsanswered written questions about the strategies theyused while nding their way and how difcult theyfound the task using each of the guidebooks Subjectsalso completed spatial and verbal neuropsychologicaltests during a separate session

62 Constants

The major independent variable in this experimentwas the type of guidebook used (text image or world-let) Parameters concerned with city design and land-mark layout as listed in table 1 were held constant

63 City Design

Our primary goal in designing the virtual citieswas to preserve as many elements of a real city as possi-ble Toward this goal we developed a pseudo-randomcity generator that produced cities based on a regular

Elvins et al 573

street grid with pavement roads and sidewalks betweenthe blocks Each block contained up to twenty build-ings side by side around the blockrsquos perimeter Using acache of 250 building designs buildings were randomlyselected and placed on city blocks Many non-landmarkbuildings were repeated between and within the citiesRendering requirements were reduced by programmingbuildings to be invisible when farther than three blocksfrom the current viewpoint To prevent subjects fromseeing buildings appearing and disappearing in the dis-tance we added fog to the environment

Cities contained parks parking lots and other civicfeatures but did not contain cars or street signs Land-marks were placed so that a subject could not short-cutthe route and would never have to make an exhaustivesearch as long as they carefully looked down each possi-ble path for familiar landmarks and landmark contextthey could always determine in which direction to pro-ceed No landmarks were located within one block ofthe perimeter wall that surrounded each city

64 Radius of Information

The radius of information was the most difcultparameter to make constant across the three guide-

books Given that text 2-D images and worldlets aredifferent representations they do not convey an equiva-lent quantity and quality of landmark information Wedetermined while building the guidebooks for the pilotstudy that using a radius of 20 m produced the mostnearly equivalent set of guidebooks For the text guide-book this meant describing in words the landmark thebuilding or structure directly across the street thebuilding on each side and whether the landmark was ona corner Thus our textual description of the landmarkshown in gure 3 read

Citgo Gas Station Citgorsquos red gas pumps are at acorner location Next door is the Cafe 360 in a beigecolored building and across the street is the LumbiniRestaurant in a red brick building

For the image guidebook a 20 m radius meant posi-tioning the camera across the street from the landmarkso that the captured thumbnail image would contain thebuildings or structures on the left and right of the land-mark For the worldlet guidebook setting the informa-tion radius simply meant setting the yon clip plane tocutaway 3-D shapes farther away than 20 m while cap-turing a spherical worldlet Image and worldlet camera

Table 1 Although the Arrangement of Buildings Landmarks and Routes Were Different in Each City Their Basic Structurein Terms of City Design and Landmark Layout was Kept Constant to Allow for Controlled Comparison Subjects Movedat a Fixed Speed in all Cities

Parameter Constant value Comments

City dimensions ten by ten blocks 100 total blocksBuilding lots per block twenty on perimeter of block building footprint variesDistance from start to goal 1200 meters past 100 12-meter-wide lotsLandmarks per city nine including goal kiosk ordered in guidebooksDistance between landmarks 15 blocks 6 025 blockTravel speed ve meterssecond slow driving speedTypes of landmarks eight one of each kind in each cityTurns required to reach goal ve left and right turnsRadius of information approximately twenty meters see city design sectionLandmark positions three on corner ve midblock goal is midblockGoal building yellowred striped kiosk six-sided viewpoint invariant

574 PRESENCE VOLUME 10 NUMBER 6

locations were selected so as not to suggest a directionof approach to the landmark

65 Hardware and Software Design

The experiment was conducted using a standard19 in monitor Stereo glasses were not used A maincity window displayed the city in the waynding VRMLbrowser Subjects used keyboard arrow keys to moveforward and backward a xed distance on each keypress and to turn left and right by a xed angle on eachkey press In the worldlet guidebook subjects used thearrow keys in the same manner to move around in theworldlet Collision detection prevented subjects frompassing through buildings and other objects in the city

Subjects pushed a Start button before beginning aLandmark button each time a landmark was reachedand a Stop button upon reaching the goal kiosk Sub-jects pushed a Guidebook button to display the guide-book Subjects practiced operating all of these buttonsin practice cities prior to beginning the experiment Thewindows were designed so that when the guidebookwas open (shown in gure 4) the main city window(shown in gure 7) was completely occluded and nomovement could be made

66 Lessons Learned from the PilotStudy

Our pilot study design and results are described byElvins et al (1997) From the pilot study we learnedthat landmarks cannot be more than two blocks apartbecause subjects frequently resort to an exhaustivesearch of the neighborhood We also learned that somesubjects will take much longer than other subjects tocomplete the task and that the task difculty and lengthmust be designed while keeping the slower wayndersin mind

67 Subject Instructions

The following is a subset of the instructions readto the subjects Instructions 3 and 4 were repeated be-fore beginning each city

1 The guidebook contains a series of landmarks thatyou probably want to follow since traveling pastthese landmarks will lead you to the goal kiosk

2 When you are at a landmark you may have towalk as far as one or two blocks before you will beable to see the next landmark

Figure 7 The main city window

Elvins et al 575

3 Your task is to locate the goal kiosk as quickly aspossible Go as fast as you can but be careful tonot get lost

4 Remember you can look at the guidebook as of-ten as you like

68 Subjects

Subjects were primarily University of CaliforniandashSan Diego students from a number of major depart-ments Ages ranged from 18 to 36 with a mean age of24 Fourteen men and twelve women completed theexperiment Results from four other subjects had to beexcluded from our data because the subjects becamelost in one or more of the cities All subjects were com-puter literate but with varying degrees of experiencewith VEs

69 Neuropsychological Tests

To provide an independent measure of a subset ofthe subjectsrsquo cognitive abilities subjects were tested in aseparate session using two standardized subtests fromthe Wechsler Adult Intelligent Scale-Revised (WAIS-R)(Wechsler 1955) The Vocabulary subtest comprisingpart of the WAIS-R Verbal subtests was used to assessthe subjectrsquos verbal ability Performance was scored ac-cording to standardized procedures based on the sub-jectrsquos degree of familiarity with a list of up to 35 wordsand their elegance of expression The second test wasthe Block Design subtest comprising part of theWAIS-R Performance subtests which was used to assessthe subjectsrsquo constructional-spatial ability and was ad-ministered in the standardized manner

610 Analysis

A one-way repeated measures analysis of variance(ANOVA) was performed for each of the dependentvariables and the overall times and distances The with-in-subjects variable was the landmark description typewith three levels text image and worldlet Post hocanalyses were completed using the Scheffe test for dif-ferences in sample means Correlation analysis was per-

formed across the scores of the neuropsychological testsand the scores of the waynding tests We adopted asignicance level of 005 on all analyses unless otherwisenoted

7 Results

Our primary result was that people spend moretime consulting the worldlet guidebook than the otherguidebooks but that they require less time to reach thegoal when using the worldlet guidebook The followingspecic ndings tell us more about the component ac-tivities that benetted from worldlet use

71 Time and Distance Measures

Figures 8 through 11 show time and distancemeasures in which the independent variable (type ofguidebook) is displayed as white gray and black barsalong with the standard error of the mean (SEM) Fig-ure 8 shows the mean overall time and distance requiredto reach the goal kiosk and gure 9 breaks the overalltime into its components Figure 10 gives the forwardand backtracking distance components Figure 11 showsthe mean number of times that each type of guidebook

Figure 8 (left) Mean total time The overall time spent traveling to

the goal was signicantly less using the worldlet guidebook than when

using the text (F(1 24) 5 989 p 001) or image (F(1 24) 5

463 p 001) guidebooks (right) Mean total distance The overall

distance traveled using the worldlet guidebook was signicantly less

than the distance traveled using the text (F(1 24) 5 1141 p

001) or image (F(1 24) 5 893 p 001) guidebooks

576 PRESENCE VOLUME 10 NUMBER 6

was opened Figure 12 diagrams the subjectsrsquo scores onthe Wechsler neuropsychological tests

Analysis of variance followed by Scheffe post hoc testswere performed for the different measures Table 2summarizes these results

72 Street Plots

Figure 13 shows plots of the streets traversed by the26 subjects in the three test cities The thickness andbrightness of each street line indicate the total number oftimes that the street was traversed For example widerbrighter streets indicate heavily used avenues Backtrackingcan increase the number of street traversals beyond theexpected minimum of 26 The starting point and goalkiosk are indicated on each plot and landmarks enroute tothe goal are marked with a star (q )

Figure 10 The mean distance moving forward while using the

worldlet guidebook is signicantly less than the forward distance

covered while using the text or image guidebook Backtracking

distance is almost nonexistent for the worldlet guidebook task

Figure 9 Turning looking for landmark is time spent standing in

one location Turning changing forward direction is time spent

changing forward direction Consulting guidebook is time spent with

the guidebook on screen Moving forward is time spent traveling over

new ground and Backtracking is time spent retreating Note that the

time spent consulting the worldlet guidebook is signicantly greater

than the time spent consulting the text or image guidebook Also note

that the time spent backtracking is near zero when using the worldlet

guidebook

Figure 11 Subjects opened the text and image guidebooks an

average of fteen to sixteen times while seeking the eight landmarks

enroute to the goal kiosk The worldlet guidebook was opened eleven

times on average

Figure 12 Z-transformed spatial and verbal scores from the

neuropsychological tests The mean raw scores were 45 and 55 for

the spatial and verbal tests respectively

Elvins et al 577

73 Subjective Measures

As part of the written questionnaire administeredafter the experiment subjects were asked how difcultthe search task was using each of the different types ofguidebooks Although responses were scattered a cleartrend in favor of the worldlet guidebook was evident asshown in table 3

74 Correlation Analysis

One surprising nding was a high inverse corre-lation ( 2 077 p 5 001) between spatial skills andthe time spent consulting the text guidebook Sub-jects with better spatial ability seemed able to form amental picture of a landmark based on textual de-scription more quickly than did subjects with poorer

spatial ability This is not the case for image andworldlet guidebooks however

We expected to nd a negative correlation betweenspatial skills and overall travel time in the image andworldlet guidebook cities The data did not supportthis nor did they support another of our hypothesesthat there would be a positive correlation between ver-bal skills and the performance in the text guidebookcity Verbal skills made no difference in overall waynd-ing time

75 Observations on Subject Strategies

Many subjects said that the tasks were harderthan they had imagined and afterwards several saidthat they felt tired No subjects complained of mo-

Table 2 Our ANOVA Indicated Signicant Differences between All the Sample Means and Post Hoc Analysis Was Warranted

ANOVA

Post hoc comparisons (Scheffe F-test)

text vs image text vs worldlet image vs worldlet

TimeConsulting guidebook 5559 p 00001 1966 p 0001 884 p 001 5488 p 0011Looking for landmark 2161 p 00001 047 1869 p 0001 1325 p 001Changing heading 1736 p 00001 002 1353 p 001 1249 p 001Moving forward 1602 p 00001 0003 1220 p 001 1183 p 001Backtracking 955 p 00003 022 826 p 001 586 p 001Overall 1034 p 00002 098 989 p 001 463 p 001

DistanceMoving forward 1533 p 00001 005 1223 p 001 1072 p 001Backtracking 1019 p 00002 023 884 p 001 621 p 001Overall 1365 p 00001 015 1141 p 001 893 p 001

GuidebookOpened 1527 p 00001 104 759 p 001 1426 p 0001

Post hoc (Scheffe) analysis indicated for all dependent variables signicant differences in the sample means of the textversus worldlet performance and image versus worldlet performance Performance using the text versus imageguidebooks was not signicantly different except that the time spent consulting the text guidebook was signicantlygreater than the time spent consulting the image guidebook In each table cell the rst number is the F value and thesecond number is the p value Asterisk indicates signicance p 005 Time and distance measures are explained in thecaptions for gures 8 through 11

578 PRESENCE VOLUME 10 NUMBER 6

tion sickness Most subjects required 70 min or moreto nish all aspects of the computer-based part of theexperiment

We observed subjects to use three general types ofstrategies while deciding what direction to go from astreet intersection

c Look down each street from many viewpointswithin the intersection until a landmark or land-mark context is seen

c Look down each street and if nothing looks famil-iar search exhaustively in each direction until thenext landmark is found

c Look down each street and if nothing looks famil-iar try going one-half block in each direction andlook again

Subjects using the rst strategy seemed to nd theirway more quickly and with less backtracking than didsubjects who used the second and third strategies An-other strategy described by some subjects on their writ-ten questionnaire involved viewing a landmark in theworldlet guidebook from many distant viewpoints in thehopes that one of the viewpoints would be similar to thedirection from which they would actually approach the

Figure 13 In these street plots line thickness and brightness

indicates the number of times that a street was traversed In top and

middle the 26 subjects traversed some streets 47 times indicating

backtracking Top and middle also exhibit circuitous trails In bottom

very little backtracking occurred and circuitous trails are minimal

Table 3 Subjects Were Asked to Rate the Difculty of theSearch Task Using the Three Different Types of Guidebooks

Text Image Worldlet

Very easy 0 2 11Easy 10 8 9Doable 10 8 6Difcult 5 7 0Very difcult 1 1 0

Median Doable Doable Easy

Elvins et al 579

landmark in the city We observed that some subjectsoften forgot what it was they were looking for and sohad to open the guidebook momentarily to remindthemselves

Not surprisingly landmarks with prominent signage(such as ldquoBobrsquos Groceryrdquo) and landmarks painted withmore highly saturated colors seemed easier to nd thanother types of landmarks Also after passing all thelandmarks several subjects exclaimed that they had lostall sense of direction within the world

A nal unexpected result were the many similarities be-tween the text and image guidebooks The text and imageguidebooks were only signicantly different in the timesubjects spent consulting them Considering that text andimages represent landmarks in very different ways we ex-pected differences in time and distance measures

76 Major Findings

In interpreting these ndings it is important toremember that the subjects performed the wayndingtasks using only the landmark knowledge gained bystudying the guidebooks No procedural or survey(map) information was provided

c As we predicted the total time spent traveling tothe goal was signicantly less using the worldletguidebook than when using the text or imageguidebooks

c Subjects spent signicantly more time studying theworldlet guidebook than they did the text or imageguidebook

c Increased worldlet study time was more than com-pensated for by a reduction in the time required toreach the goal Using the worldlet guidebook sub-jects traveled an average of 55 faster than whenusing the text guidebook and 38 faster than whenusing the image guidebook

c Increased worldlet study time also allowed subjectsto move more directly to the goal and reducedbacktracking time to almost zero This is showndiagramatically in gure 13 in which gures 13aand 13b show brighter wider streets than in gure

13c Figure 13c also exhibits fewer detours and lesswandering than do gures 13a and 13b

We postulate that compared to the landmark knowl-edge gained from text and image guidebooks the im-proved landmark knowledge built while consulting theworldlet guidebook enabled subjects to refer less fre-quently to the guidebook This postulation is furthersupported by examining worldlet guidebook results

c Subjects spent signicantly more time studying theworldlet guidebook than they did the other types ofguidebooks

c When using the worldlet guidebook subjects trav-eled faster more directly and with more skill tothe goal kiosk

c Subjectsrsquo median rating for the difculty of thesearch task was Easy when using the worldlet guide-book

8 Conclusion

The waynding literature shows that landmarksare critical in navigating an environment whether it isreal or virtual While building landmark knowledgetravelers characterize a landmark representation by its3-D shape size texture context and so forth In awaynding task textual or image guidebook landmarkdescriptions fail to engage the full range of 3-D land-mark characteristics recognized and used by travelers tond their way

We know of no studies that have considered the com-parative advantages of different types of guidebooks forwaynding Intuitively one would predict that 3-Dthumbnails make better guidebooks for 3-D worldsthan do text and 2-D thumbnail images and indeedthis is what we found We wanted however to identifythe component activities in using guidebooks that mostbenetted from 3-D elements Perhaps of greatest inter-est was our nding that worldlets enabled subjects tond their way in complex environments with almost nobacktracking By spending more time studying theworldlets in their guidebooks they spent less time look-ing around reorienting themselves conducting exhaus-

580 PRESENCE VOLUME 10 NUMBER 6

tive searches and performing similar time-wasting activ-ities We must be cautious however in what weconclude because worldlets contain more informationthan do text or 2-D thumbnail images To control forthis difference in information content more experi-ments are necessary

9 Future Work

We would like to investigate the comparative ad-vantages of worldlet guidebooks for waynding tasks incorresponding virtual and real world environments(both urban and rural) We are also considering an ex-periment to evaluate tandem use of traditional 2-Dmaps and landmark guidebooks Requiring effectivemap and compass use in a natural environment thesport of orienteering (Kjellstrom 1976) might benetfrom worldlets By encapsulating the region around theagged control points into a worldlet orienteers couldgain rst-person experience with natural landmarks suchas foliage rocks gullies shadows and so on beforebeginning the search

Acknowledgments

Our appreciation goes to Planet 9 Studios (wwwplanet9com)for the use of their city building textures Thanks to JohnMoreland and Mike Bailey for technical assistance on thisproject and thanks to Suzi Feeney for collaborating in thepilot study design UCSD librarian Debbie Cox and her staffwere invaluable in locating references Thanks also to JohnHelly and Reagan Moore for their support San Diego Super-computer Center is funded by NSF the University of Califor-nia the State of California and industrial partners

References

Appleyard D A (1976) Planning a Pluralistic City Con-icting Realities in Ciudad Guayana Cambridge MAMIT Press

Art1 Com (1997) The cybercity Berlin project [on-line]Available httpwwwartcomdeprojects

Chen S E (1995) QuickTime VRmdashAn image-based ap-proach to VE navigation Proceedings of SIGGRAPH 9529 ndash38

Darken R P amp Sibert J L (1996) Waynding strategiesand behaviors in large virtual worlds Proceedings of CHI 96142ndash149

Downs R J amp Stea D (1973) Cognitive maps and spatialbehavior In R J Downs amp D Stea (Eds) Image and En-vironment Chicago Aldine Publishing Company 8 ndash26

Elvins T T (1997) Waynding in layered informationworlds Proceedings of the CODATA Euro-American Work-shop on Data and Information Visualization 6 ndash7

Elvins T T Nadeau D R amp Kirsh D (1997) World-letsmdash3D thumbnails for waynding in virtual environmentsProceedings of the ACM User Interface Software and Technol-ogy Symposium 21ndash30

Elvins T T (1998) WorldletsmdashA novel 3D thumbnail af-fordance for waynding in large virtual environmentsUnpublished doctoral dissertation Department of Elec-trical and Computer Engineering University of Califor-niandashSan Diego

Elvins T T Nadeau D R Kirsh D amp Schul R (1998)Worldletsmdash3D thumbnails for 3D browsing Proceedings ofthe ACM CHI Conference 21ndash30

Foley J van Dam A Feiner S amp Hughes J (1990) Com-puter Graphics Principles and Practice Reading MA Addi-son-Wesley

Gillner S amp Mallot H A (1998) Navigation and acquisi-tion of spatial knowledge in a virtual maze Journal of Cog-nitive Neuroscience 10(4) 445ndash 463

Goldin S E amp Thorndyke P W (1982) Simulating naviga-tion for spatial knowledge acquisition Human Factors24(4) 457ndash 471

ISO (1997) The Virtual Reality Modeling Language Inter-national Standard ISOIEC 14772-11997 [on-line] Avail-able httpwwwvrmlorgSpecicationsVRML97

Kjellstrom B (1976) Be Expert with Map and Compass TheComplete Orienteering Handbook New York Scribner Pub-lishers

Liggett R Friedman S amp Jepson W (1995) Interactivedesigndecision making in a virtual urban world visual sim-ulation and GIS Proceedings of the 1995 ESRI User Confer-ence Environmental Systems Research [on-line] Availablehttpwwwesricombasecommonuserconfproc95

Lynch K (1960) The Image of the City Cambridge MAMIT Press

Elvins et al 581

andor themselves Simultaneously the outer mainworld is updated to match the altered miniatureautomatically adjusting the positions of shapes orthe traveler Similarly 2-D and 3-D maps are fre-quently found as navigation aids within VEs Forinstance 3-D games often provide a 3-D reduced-detail map in which an icon denotes the playerrsquoslocation Such maps can be panned zoomed androtated to provide alternate viewpoints similar tothat which is possible with miniature worlds

None of these waynding affordances offer the trav-eler faithful 3-D landmark representations or a rst-person experience with potential destinations Wickensand Baker (1995) argue that VE search tasks are dif-cult because ldquooften the object of the search is speciedin a format different from and more abstract than itsrendering in the VErdquo We address this problem by rep-resenting the object of the search in the same 3-D for-mat as the VE

3 What are Worldlets

In previous papers we described our ongoingworldlet waynding research (Elvins 1997 ElvinsNadeau amp Kirsh 1997 Elvins 1998 Elvins NadeauKirsh amp Schul 1998) Similar to a 3-D photograph aworldlet is associated with a viewing position and orien-tation and captures a snapshot of the 3-D shapes (usu-ally landmarks) that fall within the viewpointrsquos viewingvolume Worldlets can be explored and manipulated inthe same way that virtual worlds can be explored andmanipulated Incorporated into a VRML browser aguidebook or list of worldlets enables a world traveler tointeractively view landmarks and landmark context frommultiple viewpoints Seeing the landmarks presented inworldlets helps a traveler select destinations of interestand nd their way in unfamiliar worlds

In typical use the worldletrsquos viewpoint is aimed at animportant landmark and the worldletrsquos captured shapesreconstruct that landmark and its associated contextWhen viewed within an interactive 3-D browser a

worldlet provides a manipulable 3-D thumbnail repre-sentation of the landmark

Clicking on a worldlet in a landmark menu mightjump the traveler to the landmark represented by theworldlet The landmark could be at a viewpoint in thecurrent world or in another world A menu of worldletssuch as that shown in gure 4 might resemble land-mark history lists bookmarks or travel guidebooksTravelers are free to study the landmark in the worldletfrom any viewpoint before deciding whether to gothere

We have developed two types of worldlets

c A frustum worldlet contains shapes within a stan-dard pie-shaped viewing frustum positioned andoriented based upon a selected viewpoint Whenviewed a frustum worldlet looks like a pie-shapedfragment clipped from the larger main world

c A spherical worldlet contains shapes within a spheri-cal viewing bubble positioned at a selected view-point with a 360 deg eld of view When dis-played a spherical worldlet looks like a ball-shapedworld fragment similar to a snowglobe knick knackdepicting a favorite attraction

Figure 4 The navigation browserrsquos viewpoint selection window and

the waynding browserrsquos worldlet guidebook The ldquoGo tordquo button was

removed in the waynding browser

Elvins et al 569

For both worldlet types hither and yon clipping planesrestrict the extent of the worldlet ensuring that theworldlet contains a manageable subset of the largermain world Worldlet shape content is shaded and tex-tured to match the corresponding shapes in the mainworld Although the main world may have content thatchanges over time the captured worldlet remains staticrecording the content of the world at the time theworldlet was captured

Figure 3 depicts the process of constructing a spheri-cal worldlet from a fragment of a virtual city surround-ing a central viewpoint position Once captured world-lets can be browsed manipulated and explored in anumber of ways

4 Worldlets in Browsers

We have incorporated worldlets into the user in-terface for two experimental VRML browsers a naviga-tion browser and a waynding browser The navigationbrowser provides a ldquoGo tordquo feature that enables travel-ers to study select and y to world viewpoints in thecurrent world and to study select and load previouslyvisited worlds on the browserrsquos history list Guidebooksof worldlets in the waynding browser allow travelers tostudy 3-D landmark guidebooks before and duringtravel in an unfamiliar VE No go-to button is presentin this browser in which travelers manually move them-selves about based on the landmark knowledge gleanedduring study

41 Navigation Browser

Traditional VRML browsers provide a viewpointmenu that offers a choice of viewpoints each denotedby a brief textual description We have extended thisstandard feature to provide three experimental view-point selection interfaces each using worldlets All threepresent a set of worldlets one for each author-selectedviewpoint in the world A ldquoGo tordquo button ies the mainwindowrsquos viewpoint to that of the currently selectedworldlet The browser also supports on-the-y captureof worldlets from the travelerrsquos current viewpoint

c The viewpoint list window provides a list of world-lets beside a worldlet viewer Selection of a worldletfrom the list displays the worldlet in the viewerwhere it may be interactively panned zoomed androtated

c The viewpoint guidebook window presents a grid ofworldlet viewers arranged to form a guidebookphoto-album page Buttons on the window advancethe guidebook forward or back a page at a timeEach worldlet on the page can be interactively ex-amined Figure 4 shows the viewpoint guidebookwindow

c The viewpoint overlay window enables the travelerto select a worldlet from a list and overlay it atopthe main window highlighted in green This world-let overlay provides a clear indication of the world-letrsquos viewpoint position and orientation along withthe portion of the world captured within thatworldlet Figure 3b was generated using this overlaytechnique

Traditional VRML browsers provide a history list ofrecently visited worlds each denoted by its title orURL We have extended this standard feature to pro-vide two world selection interfaces each using world-lets In each world selection window a ldquoGo tordquo buttonloads into the main window the world associated withthe currently selected worldlet

c The world list window provides a list of worldletsbeside an interactive worldlet viewer similar to theviewpoint list window discussed earlier One world-let is available for each world on the browserrsquos his-tory list

c The world guidebook window uses the same guide-book photo-album layout used for the viewpointguidebook window discussed earlier One worldletis available for each world on the history list

A ldquoSave asrdquo feature of the VRML browser enables thetraveler to save a worldlet to a VRML le Using a col-lection of saved worldlets a world author can create aVRML world of worldlets Such a world acts like a 3-Ddestination index similar to a shelf full of snowglobeknick knacks When cast as a VRML anchor shape a

570 PRESENCE VOLUME 10 NUMBER 6

worldlet provides a 3-D button that when clicked onloads the associated world into the travelerrsquos browser

Figure 5 shows a world of worldlets Figure 5a shows

a close-up view of a world ldquodoorwayrdquo and a niche thatcontains a worldlet illustrating a viewpoint in thatworld Figure 5b shows a wider view of the same worldand multiple such doorways

42 Way nding Browser

The second browser was implemented specicallyfor the psychological experiment After manipulatingworldlets in a landmark guidebook to familiarize them-selves with the 3-D size shape color texture and con-text of the landmarks subjects in the experiment foundtheir way manually in the main window based upontheir landmark knowledge The landmark guidebookresembled the viewpoint selection window shown ingure 4 but it did not offer a ldquoGo tordquo button Thewaynding browser is described in detail in the methodssection

5 Implementation

Worldlets can be implemented in several waysThe implementation presented here is based on ahardware-accelerated algorithm and support algorithmsdeveloped specically for fast execution The VRMLbrowser used in this work maintains VE geometrywithin a tree-like scene graph Worldlets are also storedas scene graphs together with additional state informa-tion To capture a worldlet or to display a worldlet orVE the VRML browser traverses the associated scenegraph and feeds a 3-D graphics pipeline

51 Worldlet Capture in General

Any 3-D graphics pipeline can be roughly dividedinto two stages (i) transform clip and cull and (ii)rasterize (Foley van Dam Feiner amp Hughes 1990)The rst stage applies modeling viewing perspectiveand viewport transforms to map 3-D shapes to the 2-Dviewport Along the way shapes outside of the viewingfrustum are clipped away and backfaces are removedThe second stage uses 2-D shapes output by the rst

Figure 5 A world of worldlets that (top) associates a worldlet with

each doorway (bottom) in an environment containing multiple such

doorways Each doorway leads to a different world

Elvins et al 571

stage and draws the associated points lines and poly-gons on the screen

Worldlet capture taps into this 3-D graphics pipelineextracting the transformed shaded clipped and culledshape coordinates output by the rst stage prior to ras-terization in the second stage An extracted coordinatecontains x and y screen-space components a depth-buffer z-space component and the w coordinate Eachextracted coordinate has an associated RGB color andtexture coordinates which are computed by shadingand texture calculation phases in the rst pipeline stage

To create a worldlet these extracted coordinates areuntransformed to map them from viewport space backto world space as shown in gure 6 The inverses of the

viewport perspective viewing and modeling transformsare each applied Coordinate RGB colors and texturecoordinates are used to reconstruct 3-D worldlet geom-etry in a worldlet scene graph

Display of a worldlet passes this 3-D geometry backdown the graphics pipeline transforming clipping cull-ing and rasterizing the worldlet like any other 3-D con-tent

52 Frustum and Spherical Worldlets

A frustum worldlet is the result of capturing 3-Dgraphics pipeline output for a single traversal of thescene graph as viewed from the travelerrsquos current view-point The shape set extracted after the rst pipelinestage contains only those points lines and polygonsthat fall within the viewing frustum The worldlet con-structed by the browser from this geometry looks like apie-shaped slice cut out of the world

A spherical worldlet is the result of performing multi-ple frustum captures and combining the results TheVRML browser captures a spherical worldlet by spin-ning about a viewpoint position generating a set offrustum worldlets that each use a different viewing ori-entation Additional captures aimed straight up andstraight down complete the spherical worldlet The re-sulting set of capture geometry constructs a 360 degspherical view from the current viewpoint

When displayed the spherical worldletrsquos geometrylooks like a bubble cut out of the VE A close yon clipplane keeps the bubble small ensuring that it capturesonly landmark features in the immediate neighborhoodand not the entire virtual world

53 Worldlet Capture in OpenGL

To take advantage of the rendering speed offeredby the accelerated 3-D graphics pipeline on high-speedworkstations we implemented worldlet display and cap-ture using OpenInventor (Wernecke 1994) andOpenGL graphics libraries from Silicon Graphics Scenegraph construction and display traversal is managed byOpenInventor To capture worldlet geometry theVRML browser places the pipeline into feedback mode

Figure 6 World shapes are transformed to screen space via the

downward OpenGL graphics pipeline on the left Worldlet shapes are

inverse transformed from window coordinates to object coordinates via

the upward pipeline in the middle of the gure The intermediary

coordinate systems are named on the right

572 PRESENCE VOLUME 10 NUMBER 6

prior to a capture traversal and returns it to renderingmode following traversal

While in feedback mode the OpenGL pipeline divertsall transformed clipped and culled coordinates into abuffer provided by the browser Upon completion of acapture traversal no rasterization has taken place andthe feedback buffer contains the extracted geometry Byparsing the feedback buffer the VRML browser recon-structs worldlet geometry applying appropriate inversetransforms

OpenGL feedback buffer information includes shapecoordinates colors and texture coordinates but it doesnot include an indication of which texture image to usefor which bit of geometry To capture this additionalinformation the VRML browser uses OpenGLrsquos pass-through features to pass custom ags down through thepipeline during traversal To prepare these pass-through ags the browser augments the world scenegraph prior to traversal assigning a unique identier toeach texture image During a capture traversal eachtime a texture image is encountered the associatedidentier is passed down through the pipeline and intothe feedback buffer along with shape coordinates col-ors and texture coordinates During parsing of thefeedback buffer these texture identiers enable worldletgeometry reconstruction to apply the correct textureimages to the correct shapes

6 Methods

The psychological experiment was designed to testwhether guidebooks of worldlets are better for waynd-ing than guidebooks of text and images and if so why

61 General Description

A standard within-subject randomized designwas used Training began by reading each subject a se-ries of instructions and teaching the subject in a prac-tice virtual city (three blocks by three blocks) how tooperate the user interface Subjects were allowed tospend unlimited time in this city becoming familiar withthe controls and the online guidebooks

Each subject was then asked to nd his or her wayvia one landmark to the goal (a brightly colored kiosk)in three practice cities each ve blocks by ve blocks Inone city the guidebook represented each landmark witha paragraph of text in another city the landmarks wererepresented as 2-D thumbnail images and in a third citythe landmarks were represented as spherical worldletsThe guidebooks contained only landmark informationand subjects were given no procedural or survey infor-mation Although the image and worldlet guidebookhad the same appearance as shown in gure 4 theworldlets in the worldlet guidebook could be exploredand manipulated

After completing the practice tasks the experimentbegan and subjects were asked to nd their way to thegoal kiosk via eight ordered landmarks in three virtualcities ten blocks by ten blocks Performance in thesethree cities was timed and the subjectrsquos location orien-tation and actions were recorded at 1 sec intervalsSubjects visited the test cities in a random order tocounterbalance possible carryover effects and each ofthe cities and landmarks were different and unique

Before the computer portion of the experiment sub-jects completed a brief questionnaire on their prior ex-perience with computers After the experiment subjectsanswered written questions about the strategies theyused while nding their way and how difcult theyfound the task using each of the guidebooks Subjectsalso completed spatial and verbal neuropsychologicaltests during a separate session

62 Constants

The major independent variable in this experimentwas the type of guidebook used (text image or world-let) Parameters concerned with city design and land-mark layout as listed in table 1 were held constant

63 City Design

Our primary goal in designing the virtual citieswas to preserve as many elements of a real city as possi-ble Toward this goal we developed a pseudo-randomcity generator that produced cities based on a regular

Elvins et al 573

street grid with pavement roads and sidewalks betweenthe blocks Each block contained up to twenty build-ings side by side around the blockrsquos perimeter Using acache of 250 building designs buildings were randomlyselected and placed on city blocks Many non-landmarkbuildings were repeated between and within the citiesRendering requirements were reduced by programmingbuildings to be invisible when farther than three blocksfrom the current viewpoint To prevent subjects fromseeing buildings appearing and disappearing in the dis-tance we added fog to the environment

Cities contained parks parking lots and other civicfeatures but did not contain cars or street signs Land-marks were placed so that a subject could not short-cutthe route and would never have to make an exhaustivesearch as long as they carefully looked down each possi-ble path for familiar landmarks and landmark contextthey could always determine in which direction to pro-ceed No landmarks were located within one block ofthe perimeter wall that surrounded each city

64 Radius of Information

The radius of information was the most difcultparameter to make constant across the three guide-

books Given that text 2-D images and worldlets aredifferent representations they do not convey an equiva-lent quantity and quality of landmark information Wedetermined while building the guidebooks for the pilotstudy that using a radius of 20 m produced the mostnearly equivalent set of guidebooks For the text guide-book this meant describing in words the landmark thebuilding or structure directly across the street thebuilding on each side and whether the landmark was ona corner Thus our textual description of the landmarkshown in gure 3 read

Citgo Gas Station Citgorsquos red gas pumps are at acorner location Next door is the Cafe 360 in a beigecolored building and across the street is the LumbiniRestaurant in a red brick building

For the image guidebook a 20 m radius meant posi-tioning the camera across the street from the landmarkso that the captured thumbnail image would contain thebuildings or structures on the left and right of the land-mark For the worldlet guidebook setting the informa-tion radius simply meant setting the yon clip plane tocutaway 3-D shapes farther away than 20 m while cap-turing a spherical worldlet Image and worldlet camera

Table 1 Although the Arrangement of Buildings Landmarks and Routes Were Different in Each City Their Basic Structurein Terms of City Design and Landmark Layout was Kept Constant to Allow for Controlled Comparison Subjects Movedat a Fixed Speed in all Cities

Parameter Constant value Comments

City dimensions ten by ten blocks 100 total blocksBuilding lots per block twenty on perimeter of block building footprint variesDistance from start to goal 1200 meters past 100 12-meter-wide lotsLandmarks per city nine including goal kiosk ordered in guidebooksDistance between landmarks 15 blocks 6 025 blockTravel speed ve meterssecond slow driving speedTypes of landmarks eight one of each kind in each cityTurns required to reach goal ve left and right turnsRadius of information approximately twenty meters see city design sectionLandmark positions three on corner ve midblock goal is midblockGoal building yellowred striped kiosk six-sided viewpoint invariant

574 PRESENCE VOLUME 10 NUMBER 6

locations were selected so as not to suggest a directionof approach to the landmark

65 Hardware and Software Design

The experiment was conducted using a standard19 in monitor Stereo glasses were not used A maincity window displayed the city in the waynding VRMLbrowser Subjects used keyboard arrow keys to moveforward and backward a xed distance on each keypress and to turn left and right by a xed angle on eachkey press In the worldlet guidebook subjects used thearrow keys in the same manner to move around in theworldlet Collision detection prevented subjects frompassing through buildings and other objects in the city

Subjects pushed a Start button before beginning aLandmark button each time a landmark was reachedand a Stop button upon reaching the goal kiosk Sub-jects pushed a Guidebook button to display the guide-book Subjects practiced operating all of these buttonsin practice cities prior to beginning the experiment Thewindows were designed so that when the guidebookwas open (shown in gure 4) the main city window(shown in gure 7) was completely occluded and nomovement could be made

66 Lessons Learned from the PilotStudy

Our pilot study design and results are described byElvins et al (1997) From the pilot study we learnedthat landmarks cannot be more than two blocks apartbecause subjects frequently resort to an exhaustivesearch of the neighborhood We also learned that somesubjects will take much longer than other subjects tocomplete the task and that the task difculty and lengthmust be designed while keeping the slower wayndersin mind

67 Subject Instructions

The following is a subset of the instructions readto the subjects Instructions 3 and 4 were repeated be-fore beginning each city

1 The guidebook contains a series of landmarks thatyou probably want to follow since traveling pastthese landmarks will lead you to the goal kiosk

2 When you are at a landmark you may have towalk as far as one or two blocks before you will beable to see the next landmark

Figure 7 The main city window

Elvins et al 575

3 Your task is to locate the goal kiosk as quickly aspossible Go as fast as you can but be careful tonot get lost

4 Remember you can look at the guidebook as of-ten as you like

68 Subjects

Subjects were primarily University of CaliforniandashSan Diego students from a number of major depart-ments Ages ranged from 18 to 36 with a mean age of24 Fourteen men and twelve women completed theexperiment Results from four other subjects had to beexcluded from our data because the subjects becamelost in one or more of the cities All subjects were com-puter literate but with varying degrees of experiencewith VEs

69 Neuropsychological Tests

To provide an independent measure of a subset ofthe subjectsrsquo cognitive abilities subjects were tested in aseparate session using two standardized subtests fromthe Wechsler Adult Intelligent Scale-Revised (WAIS-R)(Wechsler 1955) The Vocabulary subtest comprisingpart of the WAIS-R Verbal subtests was used to assessthe subjectrsquos verbal ability Performance was scored ac-cording to standardized procedures based on the sub-jectrsquos degree of familiarity with a list of up to 35 wordsand their elegance of expression The second test wasthe Block Design subtest comprising part of theWAIS-R Performance subtests which was used to assessthe subjectsrsquo constructional-spatial ability and was ad-ministered in the standardized manner

610 Analysis

A one-way repeated measures analysis of variance(ANOVA) was performed for each of the dependentvariables and the overall times and distances The with-in-subjects variable was the landmark description typewith three levels text image and worldlet Post hocanalyses were completed using the Scheffe test for dif-ferences in sample means Correlation analysis was per-

formed across the scores of the neuropsychological testsand the scores of the waynding tests We adopted asignicance level of 005 on all analyses unless otherwisenoted

7 Results

Our primary result was that people spend moretime consulting the worldlet guidebook than the otherguidebooks but that they require less time to reach thegoal when using the worldlet guidebook The followingspecic ndings tell us more about the component ac-tivities that benetted from worldlet use

71 Time and Distance Measures

Figures 8 through 11 show time and distancemeasures in which the independent variable (type ofguidebook) is displayed as white gray and black barsalong with the standard error of the mean (SEM) Fig-ure 8 shows the mean overall time and distance requiredto reach the goal kiosk and gure 9 breaks the overalltime into its components Figure 10 gives the forwardand backtracking distance components Figure 11 showsthe mean number of times that each type of guidebook

Figure 8 (left) Mean total time The overall time spent traveling to

the goal was signicantly less using the worldlet guidebook than when

using the text (F(1 24) 5 989 p 001) or image (F(1 24) 5

463 p 001) guidebooks (right) Mean total distance The overall

distance traveled using the worldlet guidebook was signicantly less

than the distance traveled using the text (F(1 24) 5 1141 p

001) or image (F(1 24) 5 893 p 001) guidebooks

576 PRESENCE VOLUME 10 NUMBER 6

was opened Figure 12 diagrams the subjectsrsquo scores onthe Wechsler neuropsychological tests

Analysis of variance followed by Scheffe post hoc testswere performed for the different measures Table 2summarizes these results

72 Street Plots

Figure 13 shows plots of the streets traversed by the26 subjects in the three test cities The thickness andbrightness of each street line indicate the total number oftimes that the street was traversed For example widerbrighter streets indicate heavily used avenues Backtrackingcan increase the number of street traversals beyond theexpected minimum of 26 The starting point and goalkiosk are indicated on each plot and landmarks enroute tothe goal are marked with a star (q )

Figure 10 The mean distance moving forward while using the

worldlet guidebook is signicantly less than the forward distance

covered while using the text or image guidebook Backtracking

distance is almost nonexistent for the worldlet guidebook task

Figure 9 Turning looking for landmark is time spent standing in

one location Turning changing forward direction is time spent

changing forward direction Consulting guidebook is time spent with

the guidebook on screen Moving forward is time spent traveling over

new ground and Backtracking is time spent retreating Note that the

time spent consulting the worldlet guidebook is signicantly greater

than the time spent consulting the text or image guidebook Also note

that the time spent backtracking is near zero when using the worldlet

guidebook

Figure 11 Subjects opened the text and image guidebooks an

average of fteen to sixteen times while seeking the eight landmarks

enroute to the goal kiosk The worldlet guidebook was opened eleven

times on average

Figure 12 Z-transformed spatial and verbal scores from the

neuropsychological tests The mean raw scores were 45 and 55 for

the spatial and verbal tests respectively

Elvins et al 577

73 Subjective Measures

As part of the written questionnaire administeredafter the experiment subjects were asked how difcultthe search task was using each of the different types ofguidebooks Although responses were scattered a cleartrend in favor of the worldlet guidebook was evident asshown in table 3

74 Correlation Analysis

One surprising nding was a high inverse corre-lation ( 2 077 p 5 001) between spatial skills andthe time spent consulting the text guidebook Sub-jects with better spatial ability seemed able to form amental picture of a landmark based on textual de-scription more quickly than did subjects with poorer

spatial ability This is not the case for image andworldlet guidebooks however

We expected to nd a negative correlation betweenspatial skills and overall travel time in the image andworldlet guidebook cities The data did not supportthis nor did they support another of our hypothesesthat there would be a positive correlation between ver-bal skills and the performance in the text guidebookcity Verbal skills made no difference in overall waynd-ing time

75 Observations on Subject Strategies

Many subjects said that the tasks were harderthan they had imagined and afterwards several saidthat they felt tired No subjects complained of mo-

Table 2 Our ANOVA Indicated Signicant Differences between All the Sample Means and Post Hoc Analysis Was Warranted

ANOVA

Post hoc comparisons (Scheffe F-test)

text vs image text vs worldlet image vs worldlet

TimeConsulting guidebook 5559 p 00001 1966 p 0001 884 p 001 5488 p 0011Looking for landmark 2161 p 00001 047 1869 p 0001 1325 p 001Changing heading 1736 p 00001 002 1353 p 001 1249 p 001Moving forward 1602 p 00001 0003 1220 p 001 1183 p 001Backtracking 955 p 00003 022 826 p 001 586 p 001Overall 1034 p 00002 098 989 p 001 463 p 001

DistanceMoving forward 1533 p 00001 005 1223 p 001 1072 p 001Backtracking 1019 p 00002 023 884 p 001 621 p 001Overall 1365 p 00001 015 1141 p 001 893 p 001

GuidebookOpened 1527 p 00001 104 759 p 001 1426 p 0001

Post hoc (Scheffe) analysis indicated for all dependent variables signicant differences in the sample means of the textversus worldlet performance and image versus worldlet performance Performance using the text versus imageguidebooks was not signicantly different except that the time spent consulting the text guidebook was signicantlygreater than the time spent consulting the image guidebook In each table cell the rst number is the F value and thesecond number is the p value Asterisk indicates signicance p 005 Time and distance measures are explained in thecaptions for gures 8 through 11

578 PRESENCE VOLUME 10 NUMBER 6

tion sickness Most subjects required 70 min or moreto nish all aspects of the computer-based part of theexperiment

We observed subjects to use three general types ofstrategies while deciding what direction to go from astreet intersection

c Look down each street from many viewpointswithin the intersection until a landmark or land-mark context is seen

c Look down each street and if nothing looks famil-iar search exhaustively in each direction until thenext landmark is found

c Look down each street and if nothing looks famil-iar try going one-half block in each direction andlook again

Subjects using the rst strategy seemed to nd theirway more quickly and with less backtracking than didsubjects who used the second and third strategies An-other strategy described by some subjects on their writ-ten questionnaire involved viewing a landmark in theworldlet guidebook from many distant viewpoints in thehopes that one of the viewpoints would be similar to thedirection from which they would actually approach the

Figure 13 In these street plots line thickness and brightness

indicates the number of times that a street was traversed In top and

middle the 26 subjects traversed some streets 47 times indicating

backtracking Top and middle also exhibit circuitous trails In bottom

very little backtracking occurred and circuitous trails are minimal

Table 3 Subjects Were Asked to Rate the Difculty of theSearch Task Using the Three Different Types of Guidebooks

Text Image Worldlet

Very easy 0 2 11Easy 10 8 9Doable 10 8 6Difcult 5 7 0Very difcult 1 1 0

Median Doable Doable Easy

Elvins et al 579

landmark in the city We observed that some subjectsoften forgot what it was they were looking for and sohad to open the guidebook momentarily to remindthemselves

Not surprisingly landmarks with prominent signage(such as ldquoBobrsquos Groceryrdquo) and landmarks painted withmore highly saturated colors seemed easier to nd thanother types of landmarks Also after passing all thelandmarks several subjects exclaimed that they had lostall sense of direction within the world

A nal unexpected result were the many similarities be-tween the text and image guidebooks The text and imageguidebooks were only signicantly different in the timesubjects spent consulting them Considering that text andimages represent landmarks in very different ways we ex-pected differences in time and distance measures

76 Major Findings

In interpreting these ndings it is important toremember that the subjects performed the wayndingtasks using only the landmark knowledge gained bystudying the guidebooks No procedural or survey(map) information was provided

c As we predicted the total time spent traveling tothe goal was signicantly less using the worldletguidebook than when using the text or imageguidebooks

c Subjects spent signicantly more time studying theworldlet guidebook than they did the text or imageguidebook

c Increased worldlet study time was more than com-pensated for by a reduction in the time required toreach the goal Using the worldlet guidebook sub-jects traveled an average of 55 faster than whenusing the text guidebook and 38 faster than whenusing the image guidebook

c Increased worldlet study time also allowed subjectsto move more directly to the goal and reducedbacktracking time to almost zero This is showndiagramatically in gure 13 in which gures 13aand 13b show brighter wider streets than in gure

13c Figure 13c also exhibits fewer detours and lesswandering than do gures 13a and 13b

We postulate that compared to the landmark knowl-edge gained from text and image guidebooks the im-proved landmark knowledge built while consulting theworldlet guidebook enabled subjects to refer less fre-quently to the guidebook This postulation is furthersupported by examining worldlet guidebook results

c Subjects spent signicantly more time studying theworldlet guidebook than they did the other types ofguidebooks

c When using the worldlet guidebook subjects trav-eled faster more directly and with more skill tothe goal kiosk

c Subjectsrsquo median rating for the difculty of thesearch task was Easy when using the worldlet guide-book

8 Conclusion

The waynding literature shows that landmarksare critical in navigating an environment whether it isreal or virtual While building landmark knowledgetravelers characterize a landmark representation by its3-D shape size texture context and so forth In awaynding task textual or image guidebook landmarkdescriptions fail to engage the full range of 3-D land-mark characteristics recognized and used by travelers tond their way

We know of no studies that have considered the com-parative advantages of different types of guidebooks forwaynding Intuitively one would predict that 3-Dthumbnails make better guidebooks for 3-D worldsthan do text and 2-D thumbnail images and indeedthis is what we found We wanted however to identifythe component activities in using guidebooks that mostbenetted from 3-D elements Perhaps of greatest inter-est was our nding that worldlets enabled subjects tond their way in complex environments with almost nobacktracking By spending more time studying theworldlets in their guidebooks they spent less time look-ing around reorienting themselves conducting exhaus-

580 PRESENCE VOLUME 10 NUMBER 6

tive searches and performing similar time-wasting activ-ities We must be cautious however in what weconclude because worldlets contain more informationthan do text or 2-D thumbnail images To control forthis difference in information content more experi-ments are necessary

9 Future Work

We would like to investigate the comparative ad-vantages of worldlet guidebooks for waynding tasks incorresponding virtual and real world environments(both urban and rural) We are also considering an ex-periment to evaluate tandem use of traditional 2-Dmaps and landmark guidebooks Requiring effectivemap and compass use in a natural environment thesport of orienteering (Kjellstrom 1976) might benetfrom worldlets By encapsulating the region around theagged control points into a worldlet orienteers couldgain rst-person experience with natural landmarks suchas foliage rocks gullies shadows and so on beforebeginning the search

Acknowledgments

Our appreciation goes to Planet 9 Studios (wwwplanet9com)for the use of their city building textures Thanks to JohnMoreland and Mike Bailey for technical assistance on thisproject and thanks to Suzi Feeney for collaborating in thepilot study design UCSD librarian Debbie Cox and her staffwere invaluable in locating references Thanks also to JohnHelly and Reagan Moore for their support San Diego Super-computer Center is funded by NSF the University of Califor-nia the State of California and industrial partners

References

Appleyard D A (1976) Planning a Pluralistic City Con-icting Realities in Ciudad Guayana Cambridge MAMIT Press

Art1 Com (1997) The cybercity Berlin project [on-line]Available httpwwwartcomdeprojects

Chen S E (1995) QuickTime VRmdashAn image-based ap-proach to VE navigation Proceedings of SIGGRAPH 9529 ndash38

Darken R P amp Sibert J L (1996) Waynding strategiesand behaviors in large virtual worlds Proceedings of CHI 96142ndash149

Downs R J amp Stea D (1973) Cognitive maps and spatialbehavior In R J Downs amp D Stea (Eds) Image and En-vironment Chicago Aldine Publishing Company 8 ndash26

Elvins T T (1997) Waynding in layered informationworlds Proceedings of the CODATA Euro-American Work-shop on Data and Information Visualization 6 ndash7

Elvins T T Nadeau D R amp Kirsh D (1997) World-letsmdash3D thumbnails for waynding in virtual environmentsProceedings of the ACM User Interface Software and Technol-ogy Symposium 21ndash30

Elvins T T (1998) WorldletsmdashA novel 3D thumbnail af-fordance for waynding in large virtual environmentsUnpublished doctoral dissertation Department of Elec-trical and Computer Engineering University of Califor-niandashSan Diego

Elvins T T Nadeau D R Kirsh D amp Schul R (1998)Worldletsmdash3D thumbnails for 3D browsing Proceedings ofthe ACM CHI Conference 21ndash30

Foley J van Dam A Feiner S amp Hughes J (1990) Com-puter Graphics Principles and Practice Reading MA Addi-son-Wesley

Gillner S amp Mallot H A (1998) Navigation and acquisi-tion of spatial knowledge in a virtual maze Journal of Cog-nitive Neuroscience 10(4) 445ndash 463

Goldin S E amp Thorndyke P W (1982) Simulating naviga-tion for spatial knowledge acquisition Human Factors24(4) 457ndash 471

ISO (1997) The Virtual Reality Modeling Language Inter-national Standard ISOIEC 14772-11997 [on-line] Avail-able httpwwwvrmlorgSpecicationsVRML97

Kjellstrom B (1976) Be Expert with Map and Compass TheComplete Orienteering Handbook New York Scribner Pub-lishers

Liggett R Friedman S amp Jepson W (1995) Interactivedesigndecision making in a virtual urban world visual sim-ulation and GIS Proceedings of the 1995 ESRI User Confer-ence Environmental Systems Research [on-line] Availablehttpwwwesricombasecommonuserconfproc95

Lynch K (1960) The Image of the City Cambridge MAMIT Press

Elvins et al 581

For both worldlet types hither and yon clipping planesrestrict the extent of the worldlet ensuring that theworldlet contains a manageable subset of the largermain world Worldlet shape content is shaded and tex-tured to match the corresponding shapes in the mainworld Although the main world may have content thatchanges over time the captured worldlet remains staticrecording the content of the world at the time theworldlet was captured

Figure 3 depicts the process of constructing a spheri-cal worldlet from a fragment of a virtual city surround-ing a central viewpoint position Once captured world-lets can be browsed manipulated and explored in anumber of ways

4 Worldlets in Browsers

We have incorporated worldlets into the user in-terface for two experimental VRML browsers a naviga-tion browser and a waynding browser The navigationbrowser provides a ldquoGo tordquo feature that enables travel-ers to study select and y to world viewpoints in thecurrent world and to study select and load previouslyvisited worlds on the browserrsquos history list Guidebooksof worldlets in the waynding browser allow travelers tostudy 3-D landmark guidebooks before and duringtravel in an unfamiliar VE No go-to button is presentin this browser in which travelers manually move them-selves about based on the landmark knowledge gleanedduring study

41 Navigation Browser

Traditional VRML browsers provide a viewpointmenu that offers a choice of viewpoints each denotedby a brief textual description We have extended thisstandard feature to provide three experimental view-point selection interfaces each using worldlets All threepresent a set of worldlets one for each author-selectedviewpoint in the world A ldquoGo tordquo button ies the mainwindowrsquos viewpoint to that of the currently selectedworldlet The browser also supports on-the-y captureof worldlets from the travelerrsquos current viewpoint

c The viewpoint list window provides a list of world-lets beside a worldlet viewer Selection of a worldletfrom the list displays the worldlet in the viewerwhere it may be interactively panned zoomed androtated

c The viewpoint guidebook window presents a grid ofworldlet viewers arranged to form a guidebookphoto-album page Buttons on the window advancethe guidebook forward or back a page at a timeEach worldlet on the page can be interactively ex-amined Figure 4 shows the viewpoint guidebookwindow

c The viewpoint overlay window enables the travelerto select a worldlet from a list and overlay it atopthe main window highlighted in green This world-let overlay provides a clear indication of the world-letrsquos viewpoint position and orientation along withthe portion of the world captured within thatworldlet Figure 3b was generated using this overlaytechnique

Traditional VRML browsers provide a history list ofrecently visited worlds each denoted by its title orURL We have extended this standard feature to pro-vide two world selection interfaces each using world-lets In each world selection window a ldquoGo tordquo buttonloads into the main window the world associated withthe currently selected worldlet

c The world list window provides a list of worldletsbeside an interactive worldlet viewer similar to theviewpoint list window discussed earlier One world-let is available for each world on the browserrsquos his-tory list

c The world guidebook window uses the same guide-book photo-album layout used for the viewpointguidebook window discussed earlier One worldletis available for each world on the history list

A ldquoSave asrdquo feature of the VRML browser enables thetraveler to save a worldlet to a VRML le Using a col-lection of saved worldlets a world author can create aVRML world of worldlets Such a world acts like a 3-Ddestination index similar to a shelf full of snowglobeknick knacks When cast as a VRML anchor shape a

570 PRESENCE VOLUME 10 NUMBER 6

worldlet provides a 3-D button that when clicked onloads the associated world into the travelerrsquos browser

Figure 5 shows a world of worldlets Figure 5a shows

a close-up view of a world ldquodoorwayrdquo and a niche thatcontains a worldlet illustrating a viewpoint in thatworld Figure 5b shows a wider view of the same worldand multiple such doorways

42 Way nding Browser

The second browser was implemented specicallyfor the psychological experiment After manipulatingworldlets in a landmark guidebook to familiarize them-selves with the 3-D size shape color texture and con-text of the landmarks subjects in the experiment foundtheir way manually in the main window based upontheir landmark knowledge The landmark guidebookresembled the viewpoint selection window shown ingure 4 but it did not offer a ldquoGo tordquo button Thewaynding browser is described in detail in the methodssection

5 Implementation

Worldlets can be implemented in several waysThe implementation presented here is based on ahardware-accelerated algorithm and support algorithmsdeveloped specically for fast execution The VRMLbrowser used in this work maintains VE geometrywithin a tree-like scene graph Worldlets are also storedas scene graphs together with additional state informa-tion To capture a worldlet or to display a worldlet orVE the VRML browser traverses the associated scenegraph and feeds a 3-D graphics pipeline

51 Worldlet Capture in General

Any 3-D graphics pipeline can be roughly dividedinto two stages (i) transform clip and cull and (ii)rasterize (Foley van Dam Feiner amp Hughes 1990)The rst stage applies modeling viewing perspectiveand viewport transforms to map 3-D shapes to the 2-Dviewport Along the way shapes outside of the viewingfrustum are clipped away and backfaces are removedThe second stage uses 2-D shapes output by the rst

Figure 5 A world of worldlets that (top) associates a worldlet with

each doorway (bottom) in an environment containing multiple such

doorways Each doorway leads to a different world

Elvins et al 571

stage and draws the associated points lines and poly-gons on the screen

Worldlet capture taps into this 3-D graphics pipelineextracting the transformed shaded clipped and culledshape coordinates output by the rst stage prior to ras-terization in the second stage An extracted coordinatecontains x and y screen-space components a depth-buffer z-space component and the w coordinate Eachextracted coordinate has an associated RGB color andtexture coordinates which are computed by shadingand texture calculation phases in the rst pipeline stage

To create a worldlet these extracted coordinates areuntransformed to map them from viewport space backto world space as shown in gure 6 The inverses of the

viewport perspective viewing and modeling transformsare each applied Coordinate RGB colors and texturecoordinates are used to reconstruct 3-D worldlet geom-etry in a worldlet scene graph

Display of a worldlet passes this 3-D geometry backdown the graphics pipeline transforming clipping cull-ing and rasterizing the worldlet like any other 3-D con-tent

52 Frustum and Spherical Worldlets

A frustum worldlet is the result of capturing 3-Dgraphics pipeline output for a single traversal of thescene graph as viewed from the travelerrsquos current view-point The shape set extracted after the rst pipelinestage contains only those points lines and polygonsthat fall within the viewing frustum The worldlet con-structed by the browser from this geometry looks like apie-shaped slice cut out of the world

A spherical worldlet is the result of performing multi-ple frustum captures and combining the results TheVRML browser captures a spherical worldlet by spin-ning about a viewpoint position generating a set offrustum worldlets that each use a different viewing ori-entation Additional captures aimed straight up andstraight down complete the spherical worldlet The re-sulting set of capture geometry constructs a 360 degspherical view from the current viewpoint

When displayed the spherical worldletrsquos geometrylooks like a bubble cut out of the VE A close yon clipplane keeps the bubble small ensuring that it capturesonly landmark features in the immediate neighborhoodand not the entire virtual world

53 Worldlet Capture in OpenGL

To take advantage of the rendering speed offeredby the accelerated 3-D graphics pipeline on high-speedworkstations we implemented worldlet display and cap-ture using OpenInventor (Wernecke 1994) andOpenGL graphics libraries from Silicon Graphics Scenegraph construction and display traversal is managed byOpenInventor To capture worldlet geometry theVRML browser places the pipeline into feedback mode

Figure 6 World shapes are transformed to screen space via the

downward OpenGL graphics pipeline on the left Worldlet shapes are

inverse transformed from window coordinates to object coordinates via

the upward pipeline in the middle of the gure The intermediary

coordinate systems are named on the right

572 PRESENCE VOLUME 10 NUMBER 6

prior to a capture traversal and returns it to renderingmode following traversal

While in feedback mode the OpenGL pipeline divertsall transformed clipped and culled coordinates into abuffer provided by the browser Upon completion of acapture traversal no rasterization has taken place andthe feedback buffer contains the extracted geometry Byparsing the feedback buffer the VRML browser recon-structs worldlet geometry applying appropriate inversetransforms

OpenGL feedback buffer information includes shapecoordinates colors and texture coordinates but it doesnot include an indication of which texture image to usefor which bit of geometry To capture this additionalinformation the VRML browser uses OpenGLrsquos pass-through features to pass custom ags down through thepipeline during traversal To prepare these pass-through ags the browser augments the world scenegraph prior to traversal assigning a unique identier toeach texture image During a capture traversal eachtime a texture image is encountered the associatedidentier is passed down through the pipeline and intothe feedback buffer along with shape coordinates col-ors and texture coordinates During parsing of thefeedback buffer these texture identiers enable worldletgeometry reconstruction to apply the correct textureimages to the correct shapes

6 Methods

The psychological experiment was designed to testwhether guidebooks of worldlets are better for waynd-ing than guidebooks of text and images and if so why

61 General Description

A standard within-subject randomized designwas used Training began by reading each subject a se-ries of instructions and teaching the subject in a prac-tice virtual city (three blocks by three blocks) how tooperate the user interface Subjects were allowed tospend unlimited time in this city becoming familiar withthe controls and the online guidebooks

Each subject was then asked to nd his or her wayvia one landmark to the goal (a brightly colored kiosk)in three practice cities each ve blocks by ve blocks Inone city the guidebook represented each landmark witha paragraph of text in another city the landmarks wererepresented as 2-D thumbnail images and in a third citythe landmarks were represented as spherical worldletsThe guidebooks contained only landmark informationand subjects were given no procedural or survey infor-mation Although the image and worldlet guidebookhad the same appearance as shown in gure 4 theworldlets in the worldlet guidebook could be exploredand manipulated

After completing the practice tasks the experimentbegan and subjects were asked to nd their way to thegoal kiosk via eight ordered landmarks in three virtualcities ten blocks by ten blocks Performance in thesethree cities was timed and the subjectrsquos location orien-tation and actions were recorded at 1 sec intervalsSubjects visited the test cities in a random order tocounterbalance possible carryover effects and each ofthe cities and landmarks were different and unique

Before the computer portion of the experiment sub-jects completed a brief questionnaire on their prior ex-perience with computers After the experiment subjectsanswered written questions about the strategies theyused while nding their way and how difcult theyfound the task using each of the guidebooks Subjectsalso completed spatial and verbal neuropsychologicaltests during a separate session

62 Constants

The major independent variable in this experimentwas the type of guidebook used (text image or world-let) Parameters concerned with city design and land-mark layout as listed in table 1 were held constant

63 City Design

Our primary goal in designing the virtual citieswas to preserve as many elements of a real city as possi-ble Toward this goal we developed a pseudo-randomcity generator that produced cities based on a regular

Elvins et al 573

street grid with pavement roads and sidewalks betweenthe blocks Each block contained up to twenty build-ings side by side around the blockrsquos perimeter Using acache of 250 building designs buildings were randomlyselected and placed on city blocks Many non-landmarkbuildings were repeated between and within the citiesRendering requirements were reduced by programmingbuildings to be invisible when farther than three blocksfrom the current viewpoint To prevent subjects fromseeing buildings appearing and disappearing in the dis-tance we added fog to the environment

Cities contained parks parking lots and other civicfeatures but did not contain cars or street signs Land-marks were placed so that a subject could not short-cutthe route and would never have to make an exhaustivesearch as long as they carefully looked down each possi-ble path for familiar landmarks and landmark contextthey could always determine in which direction to pro-ceed No landmarks were located within one block ofthe perimeter wall that surrounded each city

64 Radius of Information

The radius of information was the most difcultparameter to make constant across the three guide-

books Given that text 2-D images and worldlets aredifferent representations they do not convey an equiva-lent quantity and quality of landmark information Wedetermined while building the guidebooks for the pilotstudy that using a radius of 20 m produced the mostnearly equivalent set of guidebooks For the text guide-book this meant describing in words the landmark thebuilding or structure directly across the street thebuilding on each side and whether the landmark was ona corner Thus our textual description of the landmarkshown in gure 3 read

Citgo Gas Station Citgorsquos red gas pumps are at acorner location Next door is the Cafe 360 in a beigecolored building and across the street is the LumbiniRestaurant in a red brick building

For the image guidebook a 20 m radius meant posi-tioning the camera across the street from the landmarkso that the captured thumbnail image would contain thebuildings or structures on the left and right of the land-mark For the worldlet guidebook setting the informa-tion radius simply meant setting the yon clip plane tocutaway 3-D shapes farther away than 20 m while cap-turing a spherical worldlet Image and worldlet camera

Table 1 Although the Arrangement of Buildings Landmarks and Routes Were Different in Each City Their Basic Structurein Terms of City Design and Landmark Layout was Kept Constant to Allow for Controlled Comparison Subjects Movedat a Fixed Speed in all Cities

Parameter Constant value Comments

City dimensions ten by ten blocks 100 total blocksBuilding lots per block twenty on perimeter of block building footprint variesDistance from start to goal 1200 meters past 100 12-meter-wide lotsLandmarks per city nine including goal kiosk ordered in guidebooksDistance between landmarks 15 blocks 6 025 blockTravel speed ve meterssecond slow driving speedTypes of landmarks eight one of each kind in each cityTurns required to reach goal ve left and right turnsRadius of information approximately twenty meters see city design sectionLandmark positions three on corner ve midblock goal is midblockGoal building yellowred striped kiosk six-sided viewpoint invariant

574 PRESENCE VOLUME 10 NUMBER 6

locations were selected so as not to suggest a directionof approach to the landmark

65 Hardware and Software Design

The experiment was conducted using a standard19 in monitor Stereo glasses were not used A maincity window displayed the city in the waynding VRMLbrowser Subjects used keyboard arrow keys to moveforward and backward a xed distance on each keypress and to turn left and right by a xed angle on eachkey press In the worldlet guidebook subjects used thearrow keys in the same manner to move around in theworldlet Collision detection prevented subjects frompassing through buildings and other objects in the city

Subjects pushed a Start button before beginning aLandmark button each time a landmark was reachedand a Stop button upon reaching the goal kiosk Sub-jects pushed a Guidebook button to display the guide-book Subjects practiced operating all of these buttonsin practice cities prior to beginning the experiment Thewindows were designed so that when the guidebookwas open (shown in gure 4) the main city window(shown in gure 7) was completely occluded and nomovement could be made

66 Lessons Learned from the PilotStudy

Our pilot study design and results are described byElvins et al (1997) From the pilot study we learnedthat landmarks cannot be more than two blocks apartbecause subjects frequently resort to an exhaustivesearch of the neighborhood We also learned that somesubjects will take much longer than other subjects tocomplete the task and that the task difculty and lengthmust be designed while keeping the slower wayndersin mind

67 Subject Instructions

The following is a subset of the instructions readto the subjects Instructions 3 and 4 were repeated be-fore beginning each city

1 The guidebook contains a series of landmarks thatyou probably want to follow since traveling pastthese landmarks will lead you to the goal kiosk

2 When you are at a landmark you may have towalk as far as one or two blocks before you will beable to see the next landmark

Figure 7 The main city window

Elvins et al 575

3 Your task is to locate the goal kiosk as quickly aspossible Go as fast as you can but be careful tonot get lost

4 Remember you can look at the guidebook as of-ten as you like

68 Subjects

Subjects were primarily University of CaliforniandashSan Diego students from a number of major depart-ments Ages ranged from 18 to 36 with a mean age of24 Fourteen men and twelve women completed theexperiment Results from four other subjects had to beexcluded from our data because the subjects becamelost in one or more of the cities All subjects were com-puter literate but with varying degrees of experiencewith VEs

69 Neuropsychological Tests

To provide an independent measure of a subset ofthe subjectsrsquo cognitive abilities subjects were tested in aseparate session using two standardized subtests fromthe Wechsler Adult Intelligent Scale-Revised (WAIS-R)(Wechsler 1955) The Vocabulary subtest comprisingpart of the WAIS-R Verbal subtests was used to assessthe subjectrsquos verbal ability Performance was scored ac-cording to standardized procedures based on the sub-jectrsquos degree of familiarity with a list of up to 35 wordsand their elegance of expression The second test wasthe Block Design subtest comprising part of theWAIS-R Performance subtests which was used to assessthe subjectsrsquo constructional-spatial ability and was ad-ministered in the standardized manner

610 Analysis

A one-way repeated measures analysis of variance(ANOVA) was performed for each of the dependentvariables and the overall times and distances The with-in-subjects variable was the landmark description typewith three levels text image and worldlet Post hocanalyses were completed using the Scheffe test for dif-ferences in sample means Correlation analysis was per-

formed across the scores of the neuropsychological testsand the scores of the waynding tests We adopted asignicance level of 005 on all analyses unless otherwisenoted

7 Results

Our primary result was that people spend moretime consulting the worldlet guidebook than the otherguidebooks but that they require less time to reach thegoal when using the worldlet guidebook The followingspecic ndings tell us more about the component ac-tivities that benetted from worldlet use

71 Time and Distance Measures

Figures 8 through 11 show time and distancemeasures in which the independent variable (type ofguidebook) is displayed as white gray and black barsalong with the standard error of the mean (SEM) Fig-ure 8 shows the mean overall time and distance requiredto reach the goal kiosk and gure 9 breaks the overalltime into its components Figure 10 gives the forwardand backtracking distance components Figure 11 showsthe mean number of times that each type of guidebook

Figure 8 (left) Mean total time The overall time spent traveling to

the goal was signicantly less using the worldlet guidebook than when

using the text (F(1 24) 5 989 p 001) or image (F(1 24) 5

463 p 001) guidebooks (right) Mean total distance The overall

distance traveled using the worldlet guidebook was signicantly less

than the distance traveled using the text (F(1 24) 5 1141 p

001) or image (F(1 24) 5 893 p 001) guidebooks

576 PRESENCE VOLUME 10 NUMBER 6

was opened Figure 12 diagrams the subjectsrsquo scores onthe Wechsler neuropsychological tests

Analysis of variance followed by Scheffe post hoc testswere performed for the different measures Table 2summarizes these results

72 Street Plots

Figure 13 shows plots of the streets traversed by the26 subjects in the three test cities The thickness andbrightness of each street line indicate the total number oftimes that the street was traversed For example widerbrighter streets indicate heavily used avenues Backtrackingcan increase the number of street traversals beyond theexpected minimum of 26 The starting point and goalkiosk are indicated on each plot and landmarks enroute tothe goal are marked with a star (q )

Figure 10 The mean distance moving forward while using the

worldlet guidebook is signicantly less than the forward distance

covered while using the text or image guidebook Backtracking

distance is almost nonexistent for the worldlet guidebook task

Figure 9 Turning looking for landmark is time spent standing in

one location Turning changing forward direction is time spent

changing forward direction Consulting guidebook is time spent with

the guidebook on screen Moving forward is time spent traveling over

new ground and Backtracking is time spent retreating Note that the

time spent consulting the worldlet guidebook is signicantly greater

than the time spent consulting the text or image guidebook Also note

that the time spent backtracking is near zero when using the worldlet

guidebook

Figure 11 Subjects opened the text and image guidebooks an

average of fteen to sixteen times while seeking the eight landmarks

enroute to the goal kiosk The worldlet guidebook was opened eleven

times on average

Figure 12 Z-transformed spatial and verbal scores from the

neuropsychological tests The mean raw scores were 45 and 55 for

the spatial and verbal tests respectively

Elvins et al 577

73 Subjective Measures

As part of the written questionnaire administeredafter the experiment subjects were asked how difcultthe search task was using each of the different types ofguidebooks Although responses were scattered a cleartrend in favor of the worldlet guidebook was evident asshown in table 3

74 Correlation Analysis

One surprising nding was a high inverse corre-lation ( 2 077 p 5 001) between spatial skills andthe time spent consulting the text guidebook Sub-jects with better spatial ability seemed able to form amental picture of a landmark based on textual de-scription more quickly than did subjects with poorer

spatial ability This is not the case for image andworldlet guidebooks however

We expected to nd a negative correlation betweenspatial skills and overall travel time in the image andworldlet guidebook cities The data did not supportthis nor did they support another of our hypothesesthat there would be a positive correlation between ver-bal skills and the performance in the text guidebookcity Verbal skills made no difference in overall waynd-ing time

75 Observations on Subject Strategies

Many subjects said that the tasks were harderthan they had imagined and afterwards several saidthat they felt tired No subjects complained of mo-

Table 2 Our ANOVA Indicated Signicant Differences between All the Sample Means and Post Hoc Analysis Was Warranted

ANOVA

Post hoc comparisons (Scheffe F-test)

text vs image text vs worldlet image vs worldlet

TimeConsulting guidebook 5559 p 00001 1966 p 0001 884 p 001 5488 p 0011Looking for landmark 2161 p 00001 047 1869 p 0001 1325 p 001Changing heading 1736 p 00001 002 1353 p 001 1249 p 001Moving forward 1602 p 00001 0003 1220 p 001 1183 p 001Backtracking 955 p 00003 022 826 p 001 586 p 001Overall 1034 p 00002 098 989 p 001 463 p 001

DistanceMoving forward 1533 p 00001 005 1223 p 001 1072 p 001Backtracking 1019 p 00002 023 884 p 001 621 p 001Overall 1365 p 00001 015 1141 p 001 893 p 001

GuidebookOpened 1527 p 00001 104 759 p 001 1426 p 0001

Post hoc (Scheffe) analysis indicated for all dependent variables signicant differences in the sample means of the textversus worldlet performance and image versus worldlet performance Performance using the text versus imageguidebooks was not signicantly different except that the time spent consulting the text guidebook was signicantlygreater than the time spent consulting the image guidebook In each table cell the rst number is the F value and thesecond number is the p value Asterisk indicates signicance p 005 Time and distance measures are explained in thecaptions for gures 8 through 11

578 PRESENCE VOLUME 10 NUMBER 6

tion sickness Most subjects required 70 min or moreto nish all aspects of the computer-based part of theexperiment

We observed subjects to use three general types ofstrategies while deciding what direction to go from astreet intersection

c Look down each street from many viewpointswithin the intersection until a landmark or land-mark context is seen

c Look down each street and if nothing looks famil-iar search exhaustively in each direction until thenext landmark is found

c Look down each street and if nothing looks famil-iar try going one-half block in each direction andlook again

Subjects using the rst strategy seemed to nd theirway more quickly and with less backtracking than didsubjects who used the second and third strategies An-other strategy described by some subjects on their writ-ten questionnaire involved viewing a landmark in theworldlet guidebook from many distant viewpoints in thehopes that one of the viewpoints would be similar to thedirection from which they would actually approach the

Figure 13 In these street plots line thickness and brightness

indicates the number of times that a street was traversed In top and

middle the 26 subjects traversed some streets 47 times indicating

backtracking Top and middle also exhibit circuitous trails In bottom

very little backtracking occurred and circuitous trails are minimal

Table 3 Subjects Were Asked to Rate the Difculty of theSearch Task Using the Three Different Types of Guidebooks

Text Image Worldlet

Very easy 0 2 11Easy 10 8 9Doable 10 8 6Difcult 5 7 0Very difcult 1 1 0

Median Doable Doable Easy

Elvins et al 579

landmark in the city We observed that some subjectsoften forgot what it was they were looking for and sohad to open the guidebook momentarily to remindthemselves

Not surprisingly landmarks with prominent signage(such as ldquoBobrsquos Groceryrdquo) and landmarks painted withmore highly saturated colors seemed easier to nd thanother types of landmarks Also after passing all thelandmarks several subjects exclaimed that they had lostall sense of direction within the world

A nal unexpected result were the many similarities be-tween the text and image guidebooks The text and imageguidebooks were only signicantly different in the timesubjects spent consulting them Considering that text andimages represent landmarks in very different ways we ex-pected differences in time and distance measures

76 Major Findings

In interpreting these ndings it is important toremember that the subjects performed the wayndingtasks using only the landmark knowledge gained bystudying the guidebooks No procedural or survey(map) information was provided

c As we predicted the total time spent traveling tothe goal was signicantly less using the worldletguidebook than when using the text or imageguidebooks

c Subjects spent signicantly more time studying theworldlet guidebook than they did the text or imageguidebook

c Increased worldlet study time was more than com-pensated for by a reduction in the time required toreach the goal Using the worldlet guidebook sub-jects traveled an average of 55 faster than whenusing the text guidebook and 38 faster than whenusing the image guidebook

c Increased worldlet study time also allowed subjectsto move more directly to the goal and reducedbacktracking time to almost zero This is showndiagramatically in gure 13 in which gures 13aand 13b show brighter wider streets than in gure

13c Figure 13c also exhibits fewer detours and lesswandering than do gures 13a and 13b

We postulate that compared to the landmark knowl-edge gained from text and image guidebooks the im-proved landmark knowledge built while consulting theworldlet guidebook enabled subjects to refer less fre-quently to the guidebook This postulation is furthersupported by examining worldlet guidebook results

c Subjects spent signicantly more time studying theworldlet guidebook than they did the other types ofguidebooks

c When using the worldlet guidebook subjects trav-eled faster more directly and with more skill tothe goal kiosk

c Subjectsrsquo median rating for the difculty of thesearch task was Easy when using the worldlet guide-book

8 Conclusion

The waynding literature shows that landmarksare critical in navigating an environment whether it isreal or virtual While building landmark knowledgetravelers characterize a landmark representation by its3-D shape size texture context and so forth In awaynding task textual or image guidebook landmarkdescriptions fail to engage the full range of 3-D land-mark characteristics recognized and used by travelers tond their way

We know of no studies that have considered the com-parative advantages of different types of guidebooks forwaynding Intuitively one would predict that 3-Dthumbnails make better guidebooks for 3-D worldsthan do text and 2-D thumbnail images and indeedthis is what we found We wanted however to identifythe component activities in using guidebooks that mostbenetted from 3-D elements Perhaps of greatest inter-est was our nding that worldlets enabled subjects tond their way in complex environments with almost nobacktracking By spending more time studying theworldlets in their guidebooks they spent less time look-ing around reorienting themselves conducting exhaus-

580 PRESENCE VOLUME 10 NUMBER 6

tive searches and performing similar time-wasting activ-ities We must be cautious however in what weconclude because worldlets contain more informationthan do text or 2-D thumbnail images To control forthis difference in information content more experi-ments are necessary

9 Future Work

We would like to investigate the comparative ad-vantages of worldlet guidebooks for waynding tasks incorresponding virtual and real world environments(both urban and rural) We are also considering an ex-periment to evaluate tandem use of traditional 2-Dmaps and landmark guidebooks Requiring effectivemap and compass use in a natural environment thesport of orienteering (Kjellstrom 1976) might benetfrom worldlets By encapsulating the region around theagged control points into a worldlet orienteers couldgain rst-person experience with natural landmarks suchas foliage rocks gullies shadows and so on beforebeginning the search

Acknowledgments

Our appreciation goes to Planet 9 Studios (wwwplanet9com)for the use of their city building textures Thanks to JohnMoreland and Mike Bailey for technical assistance on thisproject and thanks to Suzi Feeney for collaborating in thepilot study design UCSD librarian Debbie Cox and her staffwere invaluable in locating references Thanks also to JohnHelly and Reagan Moore for their support San Diego Super-computer Center is funded by NSF the University of Califor-nia the State of California and industrial partners

References

Appleyard D A (1976) Planning a Pluralistic City Con-icting Realities in Ciudad Guayana Cambridge MAMIT Press

Art1 Com (1997) The cybercity Berlin project [on-line]Available httpwwwartcomdeprojects

Chen S E (1995) QuickTime VRmdashAn image-based ap-proach to VE navigation Proceedings of SIGGRAPH 9529 ndash38

Darken R P amp Sibert J L (1996) Waynding strategiesand behaviors in large virtual worlds Proceedings of CHI 96142ndash149

Downs R J amp Stea D (1973) Cognitive maps and spatialbehavior In R J Downs amp D Stea (Eds) Image and En-vironment Chicago Aldine Publishing Company 8 ndash26

Elvins T T (1997) Waynding in layered informationworlds Proceedings of the CODATA Euro-American Work-shop on Data and Information Visualization 6 ndash7

Elvins T T Nadeau D R amp Kirsh D (1997) World-letsmdash3D thumbnails for waynding in virtual environmentsProceedings of the ACM User Interface Software and Technol-ogy Symposium 21ndash30

Elvins T T (1998) WorldletsmdashA novel 3D thumbnail af-fordance for waynding in large virtual environmentsUnpublished doctoral dissertation Department of Elec-trical and Computer Engineering University of Califor-niandashSan Diego

Elvins T T Nadeau D R Kirsh D amp Schul R (1998)Worldletsmdash3D thumbnails for 3D browsing Proceedings ofthe ACM CHI Conference 21ndash30

Foley J van Dam A Feiner S amp Hughes J (1990) Com-puter Graphics Principles and Practice Reading MA Addi-son-Wesley

Gillner S amp Mallot H A (1998) Navigation and acquisi-tion of spatial knowledge in a virtual maze Journal of Cog-nitive Neuroscience 10(4) 445ndash 463

Goldin S E amp Thorndyke P W (1982) Simulating naviga-tion for spatial knowledge acquisition Human Factors24(4) 457ndash 471

ISO (1997) The Virtual Reality Modeling Language Inter-national Standard ISOIEC 14772-11997 [on-line] Avail-able httpwwwvrmlorgSpecicationsVRML97

Kjellstrom B (1976) Be Expert with Map and Compass TheComplete Orienteering Handbook New York Scribner Pub-lishers

Liggett R Friedman S amp Jepson W (1995) Interactivedesigndecision making in a virtual urban world visual sim-ulation and GIS Proceedings of the 1995 ESRI User Confer-ence Environmental Systems Research [on-line] Availablehttpwwwesricombasecommonuserconfproc95

Lynch K (1960) The Image of the City Cambridge MAMIT Press

Elvins et al 581

worldlet provides a 3-D button that when clicked onloads the associated world into the travelerrsquos browser

Figure 5 shows a world of worldlets Figure 5a shows

a close-up view of a world ldquodoorwayrdquo and a niche thatcontains a worldlet illustrating a viewpoint in thatworld Figure 5b shows a wider view of the same worldand multiple such doorways

42 Way nding Browser

The second browser was implemented specicallyfor the psychological experiment After manipulatingworldlets in a landmark guidebook to familiarize them-selves with the 3-D size shape color texture and con-text of the landmarks subjects in the experiment foundtheir way manually in the main window based upontheir landmark knowledge The landmark guidebookresembled the viewpoint selection window shown ingure 4 but it did not offer a ldquoGo tordquo button Thewaynding browser is described in detail in the methodssection

5 Implementation

Worldlets can be implemented in several waysThe implementation presented here is based on ahardware-accelerated algorithm and support algorithmsdeveloped specically for fast execution The VRMLbrowser used in this work maintains VE geometrywithin a tree-like scene graph Worldlets are also storedas scene graphs together with additional state informa-tion To capture a worldlet or to display a worldlet orVE the VRML browser traverses the associated scenegraph and feeds a 3-D graphics pipeline

51 Worldlet Capture in General

Any 3-D graphics pipeline can be roughly dividedinto two stages (i) transform clip and cull and (ii)rasterize (Foley van Dam Feiner amp Hughes 1990)The rst stage applies modeling viewing perspectiveand viewport transforms to map 3-D shapes to the 2-Dviewport Along the way shapes outside of the viewingfrustum are clipped away and backfaces are removedThe second stage uses 2-D shapes output by the rst

Figure 5 A world of worldlets that (top) associates a worldlet with

each doorway (bottom) in an environment containing multiple such

doorways Each doorway leads to a different world

Elvins et al 571

stage and draws the associated points lines and poly-gons on the screen

Worldlet capture taps into this 3-D graphics pipelineextracting the transformed shaded clipped and culledshape coordinates output by the rst stage prior to ras-terization in the second stage An extracted coordinatecontains x and y screen-space components a depth-buffer z-space component and the w coordinate Eachextracted coordinate has an associated RGB color andtexture coordinates which are computed by shadingand texture calculation phases in the rst pipeline stage

To create a worldlet these extracted coordinates areuntransformed to map them from viewport space backto world space as shown in gure 6 The inverses of the

viewport perspective viewing and modeling transformsare each applied Coordinate RGB colors and texturecoordinates are used to reconstruct 3-D worldlet geom-etry in a worldlet scene graph

Display of a worldlet passes this 3-D geometry backdown the graphics pipeline transforming clipping cull-ing and rasterizing the worldlet like any other 3-D con-tent

52 Frustum and Spherical Worldlets

A frustum worldlet is the result of capturing 3-Dgraphics pipeline output for a single traversal of thescene graph as viewed from the travelerrsquos current view-point The shape set extracted after the rst pipelinestage contains only those points lines and polygonsthat fall within the viewing frustum The worldlet con-structed by the browser from this geometry looks like apie-shaped slice cut out of the world

A spherical worldlet is the result of performing multi-ple frustum captures and combining the results TheVRML browser captures a spherical worldlet by spin-ning about a viewpoint position generating a set offrustum worldlets that each use a different viewing ori-entation Additional captures aimed straight up andstraight down complete the spherical worldlet The re-sulting set of capture geometry constructs a 360 degspherical view from the current viewpoint

When displayed the spherical worldletrsquos geometrylooks like a bubble cut out of the VE A close yon clipplane keeps the bubble small ensuring that it capturesonly landmark features in the immediate neighborhoodand not the entire virtual world

53 Worldlet Capture in OpenGL

To take advantage of the rendering speed offeredby the accelerated 3-D graphics pipeline on high-speedworkstations we implemented worldlet display and cap-ture using OpenInventor (Wernecke 1994) andOpenGL graphics libraries from Silicon Graphics Scenegraph construction and display traversal is managed byOpenInventor To capture worldlet geometry theVRML browser places the pipeline into feedback mode

Figure 6 World shapes are transformed to screen space via the

downward OpenGL graphics pipeline on the left Worldlet shapes are

inverse transformed from window coordinates to object coordinates via

the upward pipeline in the middle of the gure The intermediary

coordinate systems are named on the right

572 PRESENCE VOLUME 10 NUMBER 6

prior to a capture traversal and returns it to renderingmode following traversal

While in feedback mode the OpenGL pipeline divertsall transformed clipped and culled coordinates into abuffer provided by the browser Upon completion of acapture traversal no rasterization has taken place andthe feedback buffer contains the extracted geometry Byparsing the feedback buffer the VRML browser recon-structs worldlet geometry applying appropriate inversetransforms

OpenGL feedback buffer information includes shapecoordinates colors and texture coordinates but it doesnot include an indication of which texture image to usefor which bit of geometry To capture this additionalinformation the VRML browser uses OpenGLrsquos pass-through features to pass custom ags down through thepipeline during traversal To prepare these pass-through ags the browser augments the world scenegraph prior to traversal assigning a unique identier toeach texture image During a capture traversal eachtime a texture image is encountered the associatedidentier is passed down through the pipeline and intothe feedback buffer along with shape coordinates col-ors and texture coordinates During parsing of thefeedback buffer these texture identiers enable worldletgeometry reconstruction to apply the correct textureimages to the correct shapes

6 Methods

The psychological experiment was designed to testwhether guidebooks of worldlets are better for waynd-ing than guidebooks of text and images and if so why

61 General Description

A standard within-subject randomized designwas used Training began by reading each subject a se-ries of instructions and teaching the subject in a prac-tice virtual city (three blocks by three blocks) how tooperate the user interface Subjects were allowed tospend unlimited time in this city becoming familiar withthe controls and the online guidebooks

Each subject was then asked to nd his or her wayvia one landmark to the goal (a brightly colored kiosk)in three practice cities each ve blocks by ve blocks Inone city the guidebook represented each landmark witha paragraph of text in another city the landmarks wererepresented as 2-D thumbnail images and in a third citythe landmarks were represented as spherical worldletsThe guidebooks contained only landmark informationand subjects were given no procedural or survey infor-mation Although the image and worldlet guidebookhad the same appearance as shown in gure 4 theworldlets in the worldlet guidebook could be exploredand manipulated

After completing the practice tasks the experimentbegan and subjects were asked to nd their way to thegoal kiosk via eight ordered landmarks in three virtualcities ten blocks by ten blocks Performance in thesethree cities was timed and the subjectrsquos location orien-tation and actions were recorded at 1 sec intervalsSubjects visited the test cities in a random order tocounterbalance possible carryover effects and each ofthe cities and landmarks were different and unique

Before the computer portion of the experiment sub-jects completed a brief questionnaire on their prior ex-perience with computers After the experiment subjectsanswered written questions about the strategies theyused while nding their way and how difcult theyfound the task using each of the guidebooks Subjectsalso completed spatial and verbal neuropsychologicaltests during a separate session

62 Constants

The major independent variable in this experimentwas the type of guidebook used (text image or world-let) Parameters concerned with city design and land-mark layout as listed in table 1 were held constant

63 City Design

Our primary goal in designing the virtual citieswas to preserve as many elements of a real city as possi-ble Toward this goal we developed a pseudo-randomcity generator that produced cities based on a regular

Elvins et al 573

street grid with pavement roads and sidewalks betweenthe blocks Each block contained up to twenty build-ings side by side around the blockrsquos perimeter Using acache of 250 building designs buildings were randomlyselected and placed on city blocks Many non-landmarkbuildings were repeated between and within the citiesRendering requirements were reduced by programmingbuildings to be invisible when farther than three blocksfrom the current viewpoint To prevent subjects fromseeing buildings appearing and disappearing in the dis-tance we added fog to the environment

Cities contained parks parking lots and other civicfeatures but did not contain cars or street signs Land-marks were placed so that a subject could not short-cutthe route and would never have to make an exhaustivesearch as long as they carefully looked down each possi-ble path for familiar landmarks and landmark contextthey could always determine in which direction to pro-ceed No landmarks were located within one block ofthe perimeter wall that surrounded each city

64 Radius of Information

The radius of information was the most difcultparameter to make constant across the three guide-

books Given that text 2-D images and worldlets aredifferent representations they do not convey an equiva-lent quantity and quality of landmark information Wedetermined while building the guidebooks for the pilotstudy that using a radius of 20 m produced the mostnearly equivalent set of guidebooks For the text guide-book this meant describing in words the landmark thebuilding or structure directly across the street thebuilding on each side and whether the landmark was ona corner Thus our textual description of the landmarkshown in gure 3 read

Citgo Gas Station Citgorsquos red gas pumps are at acorner location Next door is the Cafe 360 in a beigecolored building and across the street is the LumbiniRestaurant in a red brick building

For the image guidebook a 20 m radius meant posi-tioning the camera across the street from the landmarkso that the captured thumbnail image would contain thebuildings or structures on the left and right of the land-mark For the worldlet guidebook setting the informa-tion radius simply meant setting the yon clip plane tocutaway 3-D shapes farther away than 20 m while cap-turing a spherical worldlet Image and worldlet camera

Table 1 Although the Arrangement of Buildings Landmarks and Routes Were Different in Each City Their Basic Structurein Terms of City Design and Landmark Layout was Kept Constant to Allow for Controlled Comparison Subjects Movedat a Fixed Speed in all Cities

Parameter Constant value Comments

City dimensions ten by ten blocks 100 total blocksBuilding lots per block twenty on perimeter of block building footprint variesDistance from start to goal 1200 meters past 100 12-meter-wide lotsLandmarks per city nine including goal kiosk ordered in guidebooksDistance between landmarks 15 blocks 6 025 blockTravel speed ve meterssecond slow driving speedTypes of landmarks eight one of each kind in each cityTurns required to reach goal ve left and right turnsRadius of information approximately twenty meters see city design sectionLandmark positions three on corner ve midblock goal is midblockGoal building yellowred striped kiosk six-sided viewpoint invariant

574 PRESENCE VOLUME 10 NUMBER 6

locations were selected so as not to suggest a directionof approach to the landmark

65 Hardware and Software Design

The experiment was conducted using a standard19 in monitor Stereo glasses were not used A maincity window displayed the city in the waynding VRMLbrowser Subjects used keyboard arrow keys to moveforward and backward a xed distance on each keypress and to turn left and right by a xed angle on eachkey press In the worldlet guidebook subjects used thearrow keys in the same manner to move around in theworldlet Collision detection prevented subjects frompassing through buildings and other objects in the city

Subjects pushed a Start button before beginning aLandmark button each time a landmark was reachedand a Stop button upon reaching the goal kiosk Sub-jects pushed a Guidebook button to display the guide-book Subjects practiced operating all of these buttonsin practice cities prior to beginning the experiment Thewindows were designed so that when the guidebookwas open (shown in gure 4) the main city window(shown in gure 7) was completely occluded and nomovement could be made

66 Lessons Learned from the PilotStudy

Our pilot study design and results are described byElvins et al (1997) From the pilot study we learnedthat landmarks cannot be more than two blocks apartbecause subjects frequently resort to an exhaustivesearch of the neighborhood We also learned that somesubjects will take much longer than other subjects tocomplete the task and that the task difculty and lengthmust be designed while keeping the slower wayndersin mind

67 Subject Instructions

The following is a subset of the instructions readto the subjects Instructions 3 and 4 were repeated be-fore beginning each city

1 The guidebook contains a series of landmarks thatyou probably want to follow since traveling pastthese landmarks will lead you to the goal kiosk

2 When you are at a landmark you may have towalk as far as one or two blocks before you will beable to see the next landmark

Figure 7 The main city window

Elvins et al 575

3 Your task is to locate the goal kiosk as quickly aspossible Go as fast as you can but be careful tonot get lost

4 Remember you can look at the guidebook as of-ten as you like

68 Subjects

Subjects were primarily University of CaliforniandashSan Diego students from a number of major depart-ments Ages ranged from 18 to 36 with a mean age of24 Fourteen men and twelve women completed theexperiment Results from four other subjects had to beexcluded from our data because the subjects becamelost in one or more of the cities All subjects were com-puter literate but with varying degrees of experiencewith VEs

69 Neuropsychological Tests

To provide an independent measure of a subset ofthe subjectsrsquo cognitive abilities subjects were tested in aseparate session using two standardized subtests fromthe Wechsler Adult Intelligent Scale-Revised (WAIS-R)(Wechsler 1955) The Vocabulary subtest comprisingpart of the WAIS-R Verbal subtests was used to assessthe subjectrsquos verbal ability Performance was scored ac-cording to standardized procedures based on the sub-jectrsquos degree of familiarity with a list of up to 35 wordsand their elegance of expression The second test wasthe Block Design subtest comprising part of theWAIS-R Performance subtests which was used to assessthe subjectsrsquo constructional-spatial ability and was ad-ministered in the standardized manner

610 Analysis

A one-way repeated measures analysis of variance(ANOVA) was performed for each of the dependentvariables and the overall times and distances The with-in-subjects variable was the landmark description typewith three levels text image and worldlet Post hocanalyses were completed using the Scheffe test for dif-ferences in sample means Correlation analysis was per-

formed across the scores of the neuropsychological testsand the scores of the waynding tests We adopted asignicance level of 005 on all analyses unless otherwisenoted

7 Results

Our primary result was that people spend moretime consulting the worldlet guidebook than the otherguidebooks but that they require less time to reach thegoal when using the worldlet guidebook The followingspecic ndings tell us more about the component ac-tivities that benetted from worldlet use

71 Time and Distance Measures

Figures 8 through 11 show time and distancemeasures in which the independent variable (type ofguidebook) is displayed as white gray and black barsalong with the standard error of the mean (SEM) Fig-ure 8 shows the mean overall time and distance requiredto reach the goal kiosk and gure 9 breaks the overalltime into its components Figure 10 gives the forwardand backtracking distance components Figure 11 showsthe mean number of times that each type of guidebook

Figure 8 (left) Mean total time The overall time spent traveling to

the goal was signicantly less using the worldlet guidebook than when

using the text (F(1 24) 5 989 p 001) or image (F(1 24) 5

463 p 001) guidebooks (right) Mean total distance The overall

distance traveled using the worldlet guidebook was signicantly less

than the distance traveled using the text (F(1 24) 5 1141 p

001) or image (F(1 24) 5 893 p 001) guidebooks

576 PRESENCE VOLUME 10 NUMBER 6

was opened Figure 12 diagrams the subjectsrsquo scores onthe Wechsler neuropsychological tests

Analysis of variance followed by Scheffe post hoc testswere performed for the different measures Table 2summarizes these results

72 Street Plots

Figure 13 shows plots of the streets traversed by the26 subjects in the three test cities The thickness andbrightness of each street line indicate the total number oftimes that the street was traversed For example widerbrighter streets indicate heavily used avenues Backtrackingcan increase the number of street traversals beyond theexpected minimum of 26 The starting point and goalkiosk are indicated on each plot and landmarks enroute tothe goal are marked with a star (q )

Figure 10 The mean distance moving forward while using the

worldlet guidebook is signicantly less than the forward distance

covered while using the text or image guidebook Backtracking

distance is almost nonexistent for the worldlet guidebook task

Figure 9 Turning looking for landmark is time spent standing in

one location Turning changing forward direction is time spent

changing forward direction Consulting guidebook is time spent with

the guidebook on screen Moving forward is time spent traveling over

new ground and Backtracking is time spent retreating Note that the

time spent consulting the worldlet guidebook is signicantly greater

than the time spent consulting the text or image guidebook Also note

that the time spent backtracking is near zero when using the worldlet

guidebook

Figure 11 Subjects opened the text and image guidebooks an

average of fteen to sixteen times while seeking the eight landmarks

enroute to the goal kiosk The worldlet guidebook was opened eleven

times on average

Figure 12 Z-transformed spatial and verbal scores from the

neuropsychological tests The mean raw scores were 45 and 55 for

the spatial and verbal tests respectively

Elvins et al 577

73 Subjective Measures

As part of the written questionnaire administeredafter the experiment subjects were asked how difcultthe search task was using each of the different types ofguidebooks Although responses were scattered a cleartrend in favor of the worldlet guidebook was evident asshown in table 3

74 Correlation Analysis

One surprising nding was a high inverse corre-lation ( 2 077 p 5 001) between spatial skills andthe time spent consulting the text guidebook Sub-jects with better spatial ability seemed able to form amental picture of a landmark based on textual de-scription more quickly than did subjects with poorer

spatial ability This is not the case for image andworldlet guidebooks however

We expected to nd a negative correlation betweenspatial skills and overall travel time in the image andworldlet guidebook cities The data did not supportthis nor did they support another of our hypothesesthat there would be a positive correlation between ver-bal skills and the performance in the text guidebookcity Verbal skills made no difference in overall waynd-ing time

75 Observations on Subject Strategies

Many subjects said that the tasks were harderthan they had imagined and afterwards several saidthat they felt tired No subjects complained of mo-

Table 2 Our ANOVA Indicated Signicant Differences between All the Sample Means and Post Hoc Analysis Was Warranted

ANOVA

Post hoc comparisons (Scheffe F-test)

text vs image text vs worldlet image vs worldlet

TimeConsulting guidebook 5559 p 00001 1966 p 0001 884 p 001 5488 p 0011Looking for landmark 2161 p 00001 047 1869 p 0001 1325 p 001Changing heading 1736 p 00001 002 1353 p 001 1249 p 001Moving forward 1602 p 00001 0003 1220 p 001 1183 p 001Backtracking 955 p 00003 022 826 p 001 586 p 001Overall 1034 p 00002 098 989 p 001 463 p 001

DistanceMoving forward 1533 p 00001 005 1223 p 001 1072 p 001Backtracking 1019 p 00002 023 884 p 001 621 p 001Overall 1365 p 00001 015 1141 p 001 893 p 001

GuidebookOpened 1527 p 00001 104 759 p 001 1426 p 0001

Post hoc (Scheffe) analysis indicated for all dependent variables signicant differences in the sample means of the textversus worldlet performance and image versus worldlet performance Performance using the text versus imageguidebooks was not signicantly different except that the time spent consulting the text guidebook was signicantlygreater than the time spent consulting the image guidebook In each table cell the rst number is the F value and thesecond number is the p value Asterisk indicates signicance p 005 Time and distance measures are explained in thecaptions for gures 8 through 11

578 PRESENCE VOLUME 10 NUMBER 6

tion sickness Most subjects required 70 min or moreto nish all aspects of the computer-based part of theexperiment

We observed subjects to use three general types ofstrategies while deciding what direction to go from astreet intersection

c Look down each street from many viewpointswithin the intersection until a landmark or land-mark context is seen

c Look down each street and if nothing looks famil-iar search exhaustively in each direction until thenext landmark is found

c Look down each street and if nothing looks famil-iar try going one-half block in each direction andlook again

Subjects using the rst strategy seemed to nd theirway more quickly and with less backtracking than didsubjects who used the second and third strategies An-other strategy described by some subjects on their writ-ten questionnaire involved viewing a landmark in theworldlet guidebook from many distant viewpoints in thehopes that one of the viewpoints would be similar to thedirection from which they would actually approach the

Figure 13 In these street plots line thickness and brightness

indicates the number of times that a street was traversed In top and

middle the 26 subjects traversed some streets 47 times indicating

backtracking Top and middle also exhibit circuitous trails In bottom

very little backtracking occurred and circuitous trails are minimal

Table 3 Subjects Were Asked to Rate the Difculty of theSearch Task Using the Three Different Types of Guidebooks

Text Image Worldlet

Very easy 0 2 11Easy 10 8 9Doable 10 8 6Difcult 5 7 0Very difcult 1 1 0

Median Doable Doable Easy

Elvins et al 579

landmark in the city We observed that some subjectsoften forgot what it was they were looking for and sohad to open the guidebook momentarily to remindthemselves

Not surprisingly landmarks with prominent signage(such as ldquoBobrsquos Groceryrdquo) and landmarks painted withmore highly saturated colors seemed easier to nd thanother types of landmarks Also after passing all thelandmarks several subjects exclaimed that they had lostall sense of direction within the world

A nal unexpected result were the many similarities be-tween the text and image guidebooks The text and imageguidebooks were only signicantly different in the timesubjects spent consulting them Considering that text andimages represent landmarks in very different ways we ex-pected differences in time and distance measures

76 Major Findings

In interpreting these ndings it is important toremember that the subjects performed the wayndingtasks using only the landmark knowledge gained bystudying the guidebooks No procedural or survey(map) information was provided

c As we predicted the total time spent traveling tothe goal was signicantly less using the worldletguidebook than when using the text or imageguidebooks

c Subjects spent signicantly more time studying theworldlet guidebook than they did the text or imageguidebook

c Increased worldlet study time was more than com-pensated for by a reduction in the time required toreach the goal Using the worldlet guidebook sub-jects traveled an average of 55 faster than whenusing the text guidebook and 38 faster than whenusing the image guidebook

c Increased worldlet study time also allowed subjectsto move more directly to the goal and reducedbacktracking time to almost zero This is showndiagramatically in gure 13 in which gures 13aand 13b show brighter wider streets than in gure

13c Figure 13c also exhibits fewer detours and lesswandering than do gures 13a and 13b

We postulate that compared to the landmark knowl-edge gained from text and image guidebooks the im-proved landmark knowledge built while consulting theworldlet guidebook enabled subjects to refer less fre-quently to the guidebook This postulation is furthersupported by examining worldlet guidebook results

c Subjects spent signicantly more time studying theworldlet guidebook than they did the other types ofguidebooks

c When using the worldlet guidebook subjects trav-eled faster more directly and with more skill tothe goal kiosk

c Subjectsrsquo median rating for the difculty of thesearch task was Easy when using the worldlet guide-book

8 Conclusion

The waynding literature shows that landmarksare critical in navigating an environment whether it isreal or virtual While building landmark knowledgetravelers characterize a landmark representation by its3-D shape size texture context and so forth In awaynding task textual or image guidebook landmarkdescriptions fail to engage the full range of 3-D land-mark characteristics recognized and used by travelers tond their way

We know of no studies that have considered the com-parative advantages of different types of guidebooks forwaynding Intuitively one would predict that 3-Dthumbnails make better guidebooks for 3-D worldsthan do text and 2-D thumbnail images and indeedthis is what we found We wanted however to identifythe component activities in using guidebooks that mostbenetted from 3-D elements Perhaps of greatest inter-est was our nding that worldlets enabled subjects tond their way in complex environments with almost nobacktracking By spending more time studying theworldlets in their guidebooks they spent less time look-ing around reorienting themselves conducting exhaus-

580 PRESENCE VOLUME 10 NUMBER 6

tive searches and performing similar time-wasting activ-ities We must be cautious however in what weconclude because worldlets contain more informationthan do text or 2-D thumbnail images To control forthis difference in information content more experi-ments are necessary

9 Future Work

We would like to investigate the comparative ad-vantages of worldlet guidebooks for waynding tasks incorresponding virtual and real world environments(both urban and rural) We are also considering an ex-periment to evaluate tandem use of traditional 2-Dmaps and landmark guidebooks Requiring effectivemap and compass use in a natural environment thesport of orienteering (Kjellstrom 1976) might benetfrom worldlets By encapsulating the region around theagged control points into a worldlet orienteers couldgain rst-person experience with natural landmarks suchas foliage rocks gullies shadows and so on beforebeginning the search

Acknowledgments

Our appreciation goes to Planet 9 Studios (wwwplanet9com)for the use of their city building textures Thanks to JohnMoreland and Mike Bailey for technical assistance on thisproject and thanks to Suzi Feeney for collaborating in thepilot study design UCSD librarian Debbie Cox and her staffwere invaluable in locating references Thanks also to JohnHelly and Reagan Moore for their support San Diego Super-computer Center is funded by NSF the University of Califor-nia the State of California and industrial partners

References

Appleyard D A (1976) Planning a Pluralistic City Con-icting Realities in Ciudad Guayana Cambridge MAMIT Press

Art1 Com (1997) The cybercity Berlin project [on-line]Available httpwwwartcomdeprojects

Chen S E (1995) QuickTime VRmdashAn image-based ap-proach to VE navigation Proceedings of SIGGRAPH 9529 ndash38

Darken R P amp Sibert J L (1996) Waynding strategiesand behaviors in large virtual worlds Proceedings of CHI 96142ndash149

Downs R J amp Stea D (1973) Cognitive maps and spatialbehavior In R J Downs amp D Stea (Eds) Image and En-vironment Chicago Aldine Publishing Company 8 ndash26

Elvins T T (1997) Waynding in layered informationworlds Proceedings of the CODATA Euro-American Work-shop on Data and Information Visualization 6 ndash7

Elvins T T Nadeau D R amp Kirsh D (1997) World-letsmdash3D thumbnails for waynding in virtual environmentsProceedings of the ACM User Interface Software and Technol-ogy Symposium 21ndash30

Elvins T T (1998) WorldletsmdashA novel 3D thumbnail af-fordance for waynding in large virtual environmentsUnpublished doctoral dissertation Department of Elec-trical and Computer Engineering University of Califor-niandashSan Diego

Elvins T T Nadeau D R Kirsh D amp Schul R (1998)Worldletsmdash3D thumbnails for 3D browsing Proceedings ofthe ACM CHI Conference 21ndash30

Foley J van Dam A Feiner S amp Hughes J (1990) Com-puter Graphics Principles and Practice Reading MA Addi-son-Wesley

Gillner S amp Mallot H A (1998) Navigation and acquisi-tion of spatial knowledge in a virtual maze Journal of Cog-nitive Neuroscience 10(4) 445ndash 463

Goldin S E amp Thorndyke P W (1982) Simulating naviga-tion for spatial knowledge acquisition Human Factors24(4) 457ndash 471

ISO (1997) The Virtual Reality Modeling Language Inter-national Standard ISOIEC 14772-11997 [on-line] Avail-able httpwwwvrmlorgSpecicationsVRML97

Kjellstrom B (1976) Be Expert with Map and Compass TheComplete Orienteering Handbook New York Scribner Pub-lishers

Liggett R Friedman S amp Jepson W (1995) Interactivedesigndecision making in a virtual urban world visual sim-ulation and GIS Proceedings of the 1995 ESRI User Confer-ence Environmental Systems Research [on-line] Availablehttpwwwesricombasecommonuserconfproc95

Lynch K (1960) The Image of the City Cambridge MAMIT Press

Elvins et al 581

stage and draws the associated points lines and poly-gons on the screen

Worldlet capture taps into this 3-D graphics pipelineextracting the transformed shaded clipped and culledshape coordinates output by the rst stage prior to ras-terization in the second stage An extracted coordinatecontains x and y screen-space components a depth-buffer z-space component and the w coordinate Eachextracted coordinate has an associated RGB color andtexture coordinates which are computed by shadingand texture calculation phases in the rst pipeline stage

To create a worldlet these extracted coordinates areuntransformed to map them from viewport space backto world space as shown in gure 6 The inverses of the

viewport perspective viewing and modeling transformsare each applied Coordinate RGB colors and texturecoordinates are used to reconstruct 3-D worldlet geom-etry in a worldlet scene graph

Display of a worldlet passes this 3-D geometry backdown the graphics pipeline transforming clipping cull-ing and rasterizing the worldlet like any other 3-D con-tent

52 Frustum and Spherical Worldlets

A frustum worldlet is the result of capturing 3-Dgraphics pipeline output for a single traversal of thescene graph as viewed from the travelerrsquos current view-point The shape set extracted after the rst pipelinestage contains only those points lines and polygonsthat fall within the viewing frustum The worldlet con-structed by the browser from this geometry looks like apie-shaped slice cut out of the world

A spherical worldlet is the result of performing multi-ple frustum captures and combining the results TheVRML browser captures a spherical worldlet by spin-ning about a viewpoint position generating a set offrustum worldlets that each use a different viewing ori-entation Additional captures aimed straight up andstraight down complete the spherical worldlet The re-sulting set of capture geometry constructs a 360 degspherical view from the current viewpoint

When displayed the spherical worldletrsquos geometrylooks like a bubble cut out of the VE A close yon clipplane keeps the bubble small ensuring that it capturesonly landmark features in the immediate neighborhoodand not the entire virtual world

53 Worldlet Capture in OpenGL

To take advantage of the rendering speed offeredby the accelerated 3-D graphics pipeline on high-speedworkstations we implemented worldlet display and cap-ture using OpenInventor (Wernecke 1994) andOpenGL graphics libraries from Silicon Graphics Scenegraph construction and display traversal is managed byOpenInventor To capture worldlet geometry theVRML browser places the pipeline into feedback mode

Figure 6 World shapes are transformed to screen space via the

downward OpenGL graphics pipeline on the left Worldlet shapes are

inverse transformed from window coordinates to object coordinates via

the upward pipeline in the middle of the gure The intermediary

coordinate systems are named on the right

572 PRESENCE VOLUME 10 NUMBER 6

prior to a capture traversal and returns it to renderingmode following traversal

While in feedback mode the OpenGL pipeline divertsall transformed clipped and culled coordinates into abuffer provided by the browser Upon completion of acapture traversal no rasterization has taken place andthe feedback buffer contains the extracted geometry Byparsing the feedback buffer the VRML browser recon-structs worldlet geometry applying appropriate inversetransforms

OpenGL feedback buffer information includes shapecoordinates colors and texture coordinates but it doesnot include an indication of which texture image to usefor which bit of geometry To capture this additionalinformation the VRML browser uses OpenGLrsquos pass-through features to pass custom ags down through thepipeline during traversal To prepare these pass-through ags the browser augments the world scenegraph prior to traversal assigning a unique identier toeach texture image During a capture traversal eachtime a texture image is encountered the associatedidentier is passed down through the pipeline and intothe feedback buffer along with shape coordinates col-ors and texture coordinates During parsing of thefeedback buffer these texture identiers enable worldletgeometry reconstruction to apply the correct textureimages to the correct shapes

6 Methods

The psychological experiment was designed to testwhether guidebooks of worldlets are better for waynd-ing than guidebooks of text and images and if so why

61 General Description

A standard within-subject randomized designwas used Training began by reading each subject a se-ries of instructions and teaching the subject in a prac-tice virtual city (three blocks by three blocks) how tooperate the user interface Subjects were allowed tospend unlimited time in this city becoming familiar withthe controls and the online guidebooks

Each subject was then asked to nd his or her wayvia one landmark to the goal (a brightly colored kiosk)in three practice cities each ve blocks by ve blocks Inone city the guidebook represented each landmark witha paragraph of text in another city the landmarks wererepresented as 2-D thumbnail images and in a third citythe landmarks were represented as spherical worldletsThe guidebooks contained only landmark informationand subjects were given no procedural or survey infor-mation Although the image and worldlet guidebookhad the same appearance as shown in gure 4 theworldlets in the worldlet guidebook could be exploredand manipulated

After completing the practice tasks the experimentbegan and subjects were asked to nd their way to thegoal kiosk via eight ordered landmarks in three virtualcities ten blocks by ten blocks Performance in thesethree cities was timed and the subjectrsquos location orien-tation and actions were recorded at 1 sec intervalsSubjects visited the test cities in a random order tocounterbalance possible carryover effects and each ofthe cities and landmarks were different and unique

Before the computer portion of the experiment sub-jects completed a brief questionnaire on their prior ex-perience with computers After the experiment subjectsanswered written questions about the strategies theyused while nding their way and how difcult theyfound the task using each of the guidebooks Subjectsalso completed spatial and verbal neuropsychologicaltests during a separate session

62 Constants

The major independent variable in this experimentwas the type of guidebook used (text image or world-let) Parameters concerned with city design and land-mark layout as listed in table 1 were held constant

63 City Design

Our primary goal in designing the virtual citieswas to preserve as many elements of a real city as possi-ble Toward this goal we developed a pseudo-randomcity generator that produced cities based on a regular

Elvins et al 573

street grid with pavement roads and sidewalks betweenthe blocks Each block contained up to twenty build-ings side by side around the blockrsquos perimeter Using acache of 250 building designs buildings were randomlyselected and placed on city blocks Many non-landmarkbuildings were repeated between and within the citiesRendering requirements were reduced by programmingbuildings to be invisible when farther than three blocksfrom the current viewpoint To prevent subjects fromseeing buildings appearing and disappearing in the dis-tance we added fog to the environment

Cities contained parks parking lots and other civicfeatures but did not contain cars or street signs Land-marks were placed so that a subject could not short-cutthe route and would never have to make an exhaustivesearch as long as they carefully looked down each possi-ble path for familiar landmarks and landmark contextthey could always determine in which direction to pro-ceed No landmarks were located within one block ofthe perimeter wall that surrounded each city

64 Radius of Information

The radius of information was the most difcultparameter to make constant across the three guide-

books Given that text 2-D images and worldlets aredifferent representations they do not convey an equiva-lent quantity and quality of landmark information Wedetermined while building the guidebooks for the pilotstudy that using a radius of 20 m produced the mostnearly equivalent set of guidebooks For the text guide-book this meant describing in words the landmark thebuilding or structure directly across the street thebuilding on each side and whether the landmark was ona corner Thus our textual description of the landmarkshown in gure 3 read

Citgo Gas Station Citgorsquos red gas pumps are at acorner location Next door is the Cafe 360 in a beigecolored building and across the street is the LumbiniRestaurant in a red brick building

For the image guidebook a 20 m radius meant posi-tioning the camera across the street from the landmarkso that the captured thumbnail image would contain thebuildings or structures on the left and right of the land-mark For the worldlet guidebook setting the informa-tion radius simply meant setting the yon clip plane tocutaway 3-D shapes farther away than 20 m while cap-turing a spherical worldlet Image and worldlet camera

Table 1 Although the Arrangement of Buildings Landmarks and Routes Were Different in Each City Their Basic Structurein Terms of City Design and Landmark Layout was Kept Constant to Allow for Controlled Comparison Subjects Movedat a Fixed Speed in all Cities

Parameter Constant value Comments

City dimensions ten by ten blocks 100 total blocksBuilding lots per block twenty on perimeter of block building footprint variesDistance from start to goal 1200 meters past 100 12-meter-wide lotsLandmarks per city nine including goal kiosk ordered in guidebooksDistance between landmarks 15 blocks 6 025 blockTravel speed ve meterssecond slow driving speedTypes of landmarks eight one of each kind in each cityTurns required to reach goal ve left and right turnsRadius of information approximately twenty meters see city design sectionLandmark positions three on corner ve midblock goal is midblockGoal building yellowred striped kiosk six-sided viewpoint invariant

574 PRESENCE VOLUME 10 NUMBER 6

locations were selected so as not to suggest a directionof approach to the landmark

65 Hardware and Software Design

The experiment was conducted using a standard19 in monitor Stereo glasses were not used A maincity window displayed the city in the waynding VRMLbrowser Subjects used keyboard arrow keys to moveforward and backward a xed distance on each keypress and to turn left and right by a xed angle on eachkey press In the worldlet guidebook subjects used thearrow keys in the same manner to move around in theworldlet Collision detection prevented subjects frompassing through buildings and other objects in the city

Subjects pushed a Start button before beginning aLandmark button each time a landmark was reachedand a Stop button upon reaching the goal kiosk Sub-jects pushed a Guidebook button to display the guide-book Subjects practiced operating all of these buttonsin practice cities prior to beginning the experiment Thewindows were designed so that when the guidebookwas open (shown in gure 4) the main city window(shown in gure 7) was completely occluded and nomovement could be made

66 Lessons Learned from the PilotStudy

Our pilot study design and results are described byElvins et al (1997) From the pilot study we learnedthat landmarks cannot be more than two blocks apartbecause subjects frequently resort to an exhaustivesearch of the neighborhood We also learned that somesubjects will take much longer than other subjects tocomplete the task and that the task difculty and lengthmust be designed while keeping the slower wayndersin mind

67 Subject Instructions

The following is a subset of the instructions readto the subjects Instructions 3 and 4 were repeated be-fore beginning each city

1 The guidebook contains a series of landmarks thatyou probably want to follow since traveling pastthese landmarks will lead you to the goal kiosk

2 When you are at a landmark you may have towalk as far as one or two blocks before you will beable to see the next landmark

Figure 7 The main city window

Elvins et al 575

3 Your task is to locate the goal kiosk as quickly aspossible Go as fast as you can but be careful tonot get lost

4 Remember you can look at the guidebook as of-ten as you like

68 Subjects

Subjects were primarily University of CaliforniandashSan Diego students from a number of major depart-ments Ages ranged from 18 to 36 with a mean age of24 Fourteen men and twelve women completed theexperiment Results from four other subjects had to beexcluded from our data because the subjects becamelost in one or more of the cities All subjects were com-puter literate but with varying degrees of experiencewith VEs

69 Neuropsychological Tests

To provide an independent measure of a subset ofthe subjectsrsquo cognitive abilities subjects were tested in aseparate session using two standardized subtests fromthe Wechsler Adult Intelligent Scale-Revised (WAIS-R)(Wechsler 1955) The Vocabulary subtest comprisingpart of the WAIS-R Verbal subtests was used to assessthe subjectrsquos verbal ability Performance was scored ac-cording to standardized procedures based on the sub-jectrsquos degree of familiarity with a list of up to 35 wordsand their elegance of expression The second test wasthe Block Design subtest comprising part of theWAIS-R Performance subtests which was used to assessthe subjectsrsquo constructional-spatial ability and was ad-ministered in the standardized manner

610 Analysis

A one-way repeated measures analysis of variance(ANOVA) was performed for each of the dependentvariables and the overall times and distances The with-in-subjects variable was the landmark description typewith three levels text image and worldlet Post hocanalyses were completed using the Scheffe test for dif-ferences in sample means Correlation analysis was per-

formed across the scores of the neuropsychological testsand the scores of the waynding tests We adopted asignicance level of 005 on all analyses unless otherwisenoted

7 Results

Our primary result was that people spend moretime consulting the worldlet guidebook than the otherguidebooks but that they require less time to reach thegoal when using the worldlet guidebook The followingspecic ndings tell us more about the component ac-tivities that benetted from worldlet use

71 Time and Distance Measures

Figures 8 through 11 show time and distancemeasures in which the independent variable (type ofguidebook) is displayed as white gray and black barsalong with the standard error of the mean (SEM) Fig-ure 8 shows the mean overall time and distance requiredto reach the goal kiosk and gure 9 breaks the overalltime into its components Figure 10 gives the forwardand backtracking distance components Figure 11 showsthe mean number of times that each type of guidebook

Figure 8 (left) Mean total time The overall time spent traveling to

the goal was signicantly less using the worldlet guidebook than when

using the text (F(1 24) 5 989 p 001) or image (F(1 24) 5

463 p 001) guidebooks (right) Mean total distance The overall

distance traveled using the worldlet guidebook was signicantly less

than the distance traveled using the text (F(1 24) 5 1141 p

001) or image (F(1 24) 5 893 p 001) guidebooks

576 PRESENCE VOLUME 10 NUMBER 6

was opened Figure 12 diagrams the subjectsrsquo scores onthe Wechsler neuropsychological tests

Analysis of variance followed by Scheffe post hoc testswere performed for the different measures Table 2summarizes these results

72 Street Plots

Figure 13 shows plots of the streets traversed by the26 subjects in the three test cities The thickness andbrightness of each street line indicate the total number oftimes that the street was traversed For example widerbrighter streets indicate heavily used avenues Backtrackingcan increase the number of street traversals beyond theexpected minimum of 26 The starting point and goalkiosk are indicated on each plot and landmarks enroute tothe goal are marked with a star (q )

Figure 10 The mean distance moving forward while using the

worldlet guidebook is signicantly less than the forward distance

covered while using the text or image guidebook Backtracking

distance is almost nonexistent for the worldlet guidebook task

Figure 9 Turning looking for landmark is time spent standing in

one location Turning changing forward direction is time spent

changing forward direction Consulting guidebook is time spent with

the guidebook on screen Moving forward is time spent traveling over

new ground and Backtracking is time spent retreating Note that the

time spent consulting the worldlet guidebook is signicantly greater

than the time spent consulting the text or image guidebook Also note

that the time spent backtracking is near zero when using the worldlet

guidebook

Figure 11 Subjects opened the text and image guidebooks an

average of fteen to sixteen times while seeking the eight landmarks

enroute to the goal kiosk The worldlet guidebook was opened eleven

times on average

Figure 12 Z-transformed spatial and verbal scores from the

neuropsychological tests The mean raw scores were 45 and 55 for

the spatial and verbal tests respectively

Elvins et al 577

73 Subjective Measures

As part of the written questionnaire administeredafter the experiment subjects were asked how difcultthe search task was using each of the different types ofguidebooks Although responses were scattered a cleartrend in favor of the worldlet guidebook was evident asshown in table 3

74 Correlation Analysis

One surprising nding was a high inverse corre-lation ( 2 077 p 5 001) between spatial skills andthe time spent consulting the text guidebook Sub-jects with better spatial ability seemed able to form amental picture of a landmark based on textual de-scription more quickly than did subjects with poorer

spatial ability This is not the case for image andworldlet guidebooks however

We expected to nd a negative correlation betweenspatial skills and overall travel time in the image andworldlet guidebook cities The data did not supportthis nor did they support another of our hypothesesthat there would be a positive correlation between ver-bal skills and the performance in the text guidebookcity Verbal skills made no difference in overall waynd-ing time

75 Observations on Subject Strategies

Many subjects said that the tasks were harderthan they had imagined and afterwards several saidthat they felt tired No subjects complained of mo-

Table 2 Our ANOVA Indicated Signicant Differences between All the Sample Means and Post Hoc Analysis Was Warranted

ANOVA

Post hoc comparisons (Scheffe F-test)

text vs image text vs worldlet image vs worldlet

TimeConsulting guidebook 5559 p 00001 1966 p 0001 884 p 001 5488 p 0011Looking for landmark 2161 p 00001 047 1869 p 0001 1325 p 001Changing heading 1736 p 00001 002 1353 p 001 1249 p 001Moving forward 1602 p 00001 0003 1220 p 001 1183 p 001Backtracking 955 p 00003 022 826 p 001 586 p 001Overall 1034 p 00002 098 989 p 001 463 p 001

DistanceMoving forward 1533 p 00001 005 1223 p 001 1072 p 001Backtracking 1019 p 00002 023 884 p 001 621 p 001Overall 1365 p 00001 015 1141 p 001 893 p 001

GuidebookOpened 1527 p 00001 104 759 p 001 1426 p 0001

Post hoc (Scheffe) analysis indicated for all dependent variables signicant differences in the sample means of the textversus worldlet performance and image versus worldlet performance Performance using the text versus imageguidebooks was not signicantly different except that the time spent consulting the text guidebook was signicantlygreater than the time spent consulting the image guidebook In each table cell the rst number is the F value and thesecond number is the p value Asterisk indicates signicance p 005 Time and distance measures are explained in thecaptions for gures 8 through 11

578 PRESENCE VOLUME 10 NUMBER 6

tion sickness Most subjects required 70 min or moreto nish all aspects of the computer-based part of theexperiment

We observed subjects to use three general types ofstrategies while deciding what direction to go from astreet intersection

c Look down each street from many viewpointswithin the intersection until a landmark or land-mark context is seen

c Look down each street and if nothing looks famil-iar search exhaustively in each direction until thenext landmark is found

c Look down each street and if nothing looks famil-iar try going one-half block in each direction andlook again

Subjects using the rst strategy seemed to nd theirway more quickly and with less backtracking than didsubjects who used the second and third strategies An-other strategy described by some subjects on their writ-ten questionnaire involved viewing a landmark in theworldlet guidebook from many distant viewpoints in thehopes that one of the viewpoints would be similar to thedirection from which they would actually approach the

Figure 13 In these street plots line thickness and brightness

indicates the number of times that a street was traversed In top and

middle the 26 subjects traversed some streets 47 times indicating

backtracking Top and middle also exhibit circuitous trails In bottom

very little backtracking occurred and circuitous trails are minimal

Table 3 Subjects Were Asked to Rate the Difculty of theSearch Task Using the Three Different Types of Guidebooks

Text Image Worldlet

Very easy 0 2 11Easy 10 8 9Doable 10 8 6Difcult 5 7 0Very difcult 1 1 0

Median Doable Doable Easy

Elvins et al 579

landmark in the city We observed that some subjectsoften forgot what it was they were looking for and sohad to open the guidebook momentarily to remindthemselves

Not surprisingly landmarks with prominent signage(such as ldquoBobrsquos Groceryrdquo) and landmarks painted withmore highly saturated colors seemed easier to nd thanother types of landmarks Also after passing all thelandmarks several subjects exclaimed that they had lostall sense of direction within the world

A nal unexpected result were the many similarities be-tween the text and image guidebooks The text and imageguidebooks were only signicantly different in the timesubjects spent consulting them Considering that text andimages represent landmarks in very different ways we ex-pected differences in time and distance measures

76 Major Findings

In interpreting these ndings it is important toremember that the subjects performed the wayndingtasks using only the landmark knowledge gained bystudying the guidebooks No procedural or survey(map) information was provided

c As we predicted the total time spent traveling tothe goal was signicantly less using the worldletguidebook than when using the text or imageguidebooks

c Subjects spent signicantly more time studying theworldlet guidebook than they did the text or imageguidebook

c Increased worldlet study time was more than com-pensated for by a reduction in the time required toreach the goal Using the worldlet guidebook sub-jects traveled an average of 55 faster than whenusing the text guidebook and 38 faster than whenusing the image guidebook

c Increased worldlet study time also allowed subjectsto move more directly to the goal and reducedbacktracking time to almost zero This is showndiagramatically in gure 13 in which gures 13aand 13b show brighter wider streets than in gure

13c Figure 13c also exhibits fewer detours and lesswandering than do gures 13a and 13b

We postulate that compared to the landmark knowl-edge gained from text and image guidebooks the im-proved landmark knowledge built while consulting theworldlet guidebook enabled subjects to refer less fre-quently to the guidebook This postulation is furthersupported by examining worldlet guidebook results

c Subjects spent signicantly more time studying theworldlet guidebook than they did the other types ofguidebooks

c When using the worldlet guidebook subjects trav-eled faster more directly and with more skill tothe goal kiosk

c Subjectsrsquo median rating for the difculty of thesearch task was Easy when using the worldlet guide-book

8 Conclusion

The waynding literature shows that landmarksare critical in navigating an environment whether it isreal or virtual While building landmark knowledgetravelers characterize a landmark representation by its3-D shape size texture context and so forth In awaynding task textual or image guidebook landmarkdescriptions fail to engage the full range of 3-D land-mark characteristics recognized and used by travelers tond their way

We know of no studies that have considered the com-parative advantages of different types of guidebooks forwaynding Intuitively one would predict that 3-Dthumbnails make better guidebooks for 3-D worldsthan do text and 2-D thumbnail images and indeedthis is what we found We wanted however to identifythe component activities in using guidebooks that mostbenetted from 3-D elements Perhaps of greatest inter-est was our nding that worldlets enabled subjects tond their way in complex environments with almost nobacktracking By spending more time studying theworldlets in their guidebooks they spent less time look-ing around reorienting themselves conducting exhaus-

580 PRESENCE VOLUME 10 NUMBER 6

tive searches and performing similar time-wasting activ-ities We must be cautious however in what weconclude because worldlets contain more informationthan do text or 2-D thumbnail images To control forthis difference in information content more experi-ments are necessary

9 Future Work

We would like to investigate the comparative ad-vantages of worldlet guidebooks for waynding tasks incorresponding virtual and real world environments(both urban and rural) We are also considering an ex-periment to evaluate tandem use of traditional 2-Dmaps and landmark guidebooks Requiring effectivemap and compass use in a natural environment thesport of orienteering (Kjellstrom 1976) might benetfrom worldlets By encapsulating the region around theagged control points into a worldlet orienteers couldgain rst-person experience with natural landmarks suchas foliage rocks gullies shadows and so on beforebeginning the search

Acknowledgments

Our appreciation goes to Planet 9 Studios (wwwplanet9com)for the use of their city building textures Thanks to JohnMoreland and Mike Bailey for technical assistance on thisproject and thanks to Suzi Feeney for collaborating in thepilot study design UCSD librarian Debbie Cox and her staffwere invaluable in locating references Thanks also to JohnHelly and Reagan Moore for their support San Diego Super-computer Center is funded by NSF the University of Califor-nia the State of California and industrial partners

References

Appleyard D A (1976) Planning a Pluralistic City Con-icting Realities in Ciudad Guayana Cambridge MAMIT Press

Art1 Com (1997) The cybercity Berlin project [on-line]Available httpwwwartcomdeprojects

Chen S E (1995) QuickTime VRmdashAn image-based ap-proach to VE navigation Proceedings of SIGGRAPH 9529 ndash38

Darken R P amp Sibert J L (1996) Waynding strategiesand behaviors in large virtual worlds Proceedings of CHI 96142ndash149

Downs R J amp Stea D (1973) Cognitive maps and spatialbehavior In R J Downs amp D Stea (Eds) Image and En-vironment Chicago Aldine Publishing Company 8 ndash26

Elvins T T (1997) Waynding in layered informationworlds Proceedings of the CODATA Euro-American Work-shop on Data and Information Visualization 6 ndash7

Elvins T T Nadeau D R amp Kirsh D (1997) World-letsmdash3D thumbnails for waynding in virtual environmentsProceedings of the ACM User Interface Software and Technol-ogy Symposium 21ndash30

Elvins T T (1998) WorldletsmdashA novel 3D thumbnail af-fordance for waynding in large virtual environmentsUnpublished doctoral dissertation Department of Elec-trical and Computer Engineering University of Califor-niandashSan Diego

Elvins T T Nadeau D R Kirsh D amp Schul R (1998)Worldletsmdash3D thumbnails for 3D browsing Proceedings ofthe ACM CHI Conference 21ndash30

Foley J van Dam A Feiner S amp Hughes J (1990) Com-puter Graphics Principles and Practice Reading MA Addi-son-Wesley

Gillner S amp Mallot H A (1998) Navigation and acquisi-tion of spatial knowledge in a virtual maze Journal of Cog-nitive Neuroscience 10(4) 445ndash 463

Goldin S E amp Thorndyke P W (1982) Simulating naviga-tion for spatial knowledge acquisition Human Factors24(4) 457ndash 471

ISO (1997) The Virtual Reality Modeling Language Inter-national Standard ISOIEC 14772-11997 [on-line] Avail-able httpwwwvrmlorgSpecicationsVRML97

Kjellstrom B (1976) Be Expert with Map and Compass TheComplete Orienteering Handbook New York Scribner Pub-lishers

Liggett R Friedman S amp Jepson W (1995) Interactivedesigndecision making in a virtual urban world visual sim-ulation and GIS Proceedings of the 1995 ESRI User Confer-ence Environmental Systems Research [on-line] Availablehttpwwwesricombasecommonuserconfproc95

Lynch K (1960) The Image of the City Cambridge MAMIT Press

Elvins et al 581

prior to a capture traversal and returns it to renderingmode following traversal

While in feedback mode the OpenGL pipeline divertsall transformed clipped and culled coordinates into abuffer provided by the browser Upon completion of acapture traversal no rasterization has taken place andthe feedback buffer contains the extracted geometry Byparsing the feedback buffer the VRML browser recon-structs worldlet geometry applying appropriate inversetransforms

OpenGL feedback buffer information includes shapecoordinates colors and texture coordinates but it doesnot include an indication of which texture image to usefor which bit of geometry To capture this additionalinformation the VRML browser uses OpenGLrsquos pass-through features to pass custom ags down through thepipeline during traversal To prepare these pass-through ags the browser augments the world scenegraph prior to traversal assigning a unique identier toeach texture image During a capture traversal eachtime a texture image is encountered the associatedidentier is passed down through the pipeline and intothe feedback buffer along with shape coordinates col-ors and texture coordinates During parsing of thefeedback buffer these texture identiers enable worldletgeometry reconstruction to apply the correct textureimages to the correct shapes

6 Methods

The psychological experiment was designed to testwhether guidebooks of worldlets are better for waynd-ing than guidebooks of text and images and if so why

61 General Description

A standard within-subject randomized designwas used Training began by reading each subject a se-ries of instructions and teaching the subject in a prac-tice virtual city (three blocks by three blocks) how tooperate the user interface Subjects were allowed tospend unlimited time in this city becoming familiar withthe controls and the online guidebooks

Each subject was then asked to nd his or her wayvia one landmark to the goal (a brightly colored kiosk)in three practice cities each ve blocks by ve blocks Inone city the guidebook represented each landmark witha paragraph of text in another city the landmarks wererepresented as 2-D thumbnail images and in a third citythe landmarks were represented as spherical worldletsThe guidebooks contained only landmark informationand subjects were given no procedural or survey infor-mation Although the image and worldlet guidebookhad the same appearance as shown in gure 4 theworldlets in the worldlet guidebook could be exploredand manipulated

After completing the practice tasks the experimentbegan and subjects were asked to nd their way to thegoal kiosk via eight ordered landmarks in three virtualcities ten blocks by ten blocks Performance in thesethree cities was timed and the subjectrsquos location orien-tation and actions were recorded at 1 sec intervalsSubjects visited the test cities in a random order tocounterbalance possible carryover effects and each ofthe cities and landmarks were different and unique

Before the computer portion of the experiment sub-jects completed a brief questionnaire on their prior ex-perience with computers After the experiment subjectsanswered written questions about the strategies theyused while nding their way and how difcult theyfound the task using each of the guidebooks Subjectsalso completed spatial and verbal neuropsychologicaltests during a separate session

62 Constants

The major independent variable in this experimentwas the type of guidebook used (text image or world-let) Parameters concerned with city design and land-mark layout as listed in table 1 were held constant

63 City Design

Our primary goal in designing the virtual citieswas to preserve as many elements of a real city as possi-ble Toward this goal we developed a pseudo-randomcity generator that produced cities based on a regular

Elvins et al 573

street grid with pavement roads and sidewalks betweenthe blocks Each block contained up to twenty build-ings side by side around the blockrsquos perimeter Using acache of 250 building designs buildings were randomlyselected and placed on city blocks Many non-landmarkbuildings were repeated between and within the citiesRendering requirements were reduced by programmingbuildings to be invisible when farther than three blocksfrom the current viewpoint To prevent subjects fromseeing buildings appearing and disappearing in the dis-tance we added fog to the environment

Cities contained parks parking lots and other civicfeatures but did not contain cars or street signs Land-marks were placed so that a subject could not short-cutthe route and would never have to make an exhaustivesearch as long as they carefully looked down each possi-ble path for familiar landmarks and landmark contextthey could always determine in which direction to pro-ceed No landmarks were located within one block ofthe perimeter wall that surrounded each city

64 Radius of Information

The radius of information was the most difcultparameter to make constant across the three guide-

books Given that text 2-D images and worldlets aredifferent representations they do not convey an equiva-lent quantity and quality of landmark information Wedetermined while building the guidebooks for the pilotstudy that using a radius of 20 m produced the mostnearly equivalent set of guidebooks For the text guide-book this meant describing in words the landmark thebuilding or structure directly across the street thebuilding on each side and whether the landmark was ona corner Thus our textual description of the landmarkshown in gure 3 read

Citgo Gas Station Citgorsquos red gas pumps are at acorner location Next door is the Cafe 360 in a beigecolored building and across the street is the LumbiniRestaurant in a red brick building

For the image guidebook a 20 m radius meant posi-tioning the camera across the street from the landmarkso that the captured thumbnail image would contain thebuildings or structures on the left and right of the land-mark For the worldlet guidebook setting the informa-tion radius simply meant setting the yon clip plane tocutaway 3-D shapes farther away than 20 m while cap-turing a spherical worldlet Image and worldlet camera

Table 1 Although the Arrangement of Buildings Landmarks and Routes Were Different in Each City Their Basic Structurein Terms of City Design and Landmark Layout was Kept Constant to Allow for Controlled Comparison Subjects Movedat a Fixed Speed in all Cities

Parameter Constant value Comments

City dimensions ten by ten blocks 100 total blocksBuilding lots per block twenty on perimeter of block building footprint variesDistance from start to goal 1200 meters past 100 12-meter-wide lotsLandmarks per city nine including goal kiosk ordered in guidebooksDistance between landmarks 15 blocks 6 025 blockTravel speed ve meterssecond slow driving speedTypes of landmarks eight one of each kind in each cityTurns required to reach goal ve left and right turnsRadius of information approximately twenty meters see city design sectionLandmark positions three on corner ve midblock goal is midblockGoal building yellowred striped kiosk six-sided viewpoint invariant

574 PRESENCE VOLUME 10 NUMBER 6

locations were selected so as not to suggest a directionof approach to the landmark

65 Hardware and Software Design

The experiment was conducted using a standard19 in monitor Stereo glasses were not used A maincity window displayed the city in the waynding VRMLbrowser Subjects used keyboard arrow keys to moveforward and backward a xed distance on each keypress and to turn left and right by a xed angle on eachkey press In the worldlet guidebook subjects used thearrow keys in the same manner to move around in theworldlet Collision detection prevented subjects frompassing through buildings and other objects in the city

Subjects pushed a Start button before beginning aLandmark button each time a landmark was reachedand a Stop button upon reaching the goal kiosk Sub-jects pushed a Guidebook button to display the guide-book Subjects practiced operating all of these buttonsin practice cities prior to beginning the experiment Thewindows were designed so that when the guidebookwas open (shown in gure 4) the main city window(shown in gure 7) was completely occluded and nomovement could be made

66 Lessons Learned from the PilotStudy

Our pilot study design and results are described byElvins et al (1997) From the pilot study we learnedthat landmarks cannot be more than two blocks apartbecause subjects frequently resort to an exhaustivesearch of the neighborhood We also learned that somesubjects will take much longer than other subjects tocomplete the task and that the task difculty and lengthmust be designed while keeping the slower wayndersin mind

67 Subject Instructions

The following is a subset of the instructions readto the subjects Instructions 3 and 4 were repeated be-fore beginning each city

1 The guidebook contains a series of landmarks thatyou probably want to follow since traveling pastthese landmarks will lead you to the goal kiosk

2 When you are at a landmark you may have towalk as far as one or two blocks before you will beable to see the next landmark

Figure 7 The main city window

Elvins et al 575

3 Your task is to locate the goal kiosk as quickly aspossible Go as fast as you can but be careful tonot get lost

4 Remember you can look at the guidebook as of-ten as you like

68 Subjects

Subjects were primarily University of CaliforniandashSan Diego students from a number of major depart-ments Ages ranged from 18 to 36 with a mean age of24 Fourteen men and twelve women completed theexperiment Results from four other subjects had to beexcluded from our data because the subjects becamelost in one or more of the cities All subjects were com-puter literate but with varying degrees of experiencewith VEs

69 Neuropsychological Tests

To provide an independent measure of a subset ofthe subjectsrsquo cognitive abilities subjects were tested in aseparate session using two standardized subtests fromthe Wechsler Adult Intelligent Scale-Revised (WAIS-R)(Wechsler 1955) The Vocabulary subtest comprisingpart of the WAIS-R Verbal subtests was used to assessthe subjectrsquos verbal ability Performance was scored ac-cording to standardized procedures based on the sub-jectrsquos degree of familiarity with a list of up to 35 wordsand their elegance of expression The second test wasthe Block Design subtest comprising part of theWAIS-R Performance subtests which was used to assessthe subjectsrsquo constructional-spatial ability and was ad-ministered in the standardized manner

610 Analysis

A one-way repeated measures analysis of variance(ANOVA) was performed for each of the dependentvariables and the overall times and distances The with-in-subjects variable was the landmark description typewith three levels text image and worldlet Post hocanalyses were completed using the Scheffe test for dif-ferences in sample means Correlation analysis was per-

formed across the scores of the neuropsychological testsand the scores of the waynding tests We adopted asignicance level of 005 on all analyses unless otherwisenoted

7 Results

Our primary result was that people spend moretime consulting the worldlet guidebook than the otherguidebooks but that they require less time to reach thegoal when using the worldlet guidebook The followingspecic ndings tell us more about the component ac-tivities that benetted from worldlet use

71 Time and Distance Measures

Figures 8 through 11 show time and distancemeasures in which the independent variable (type ofguidebook) is displayed as white gray and black barsalong with the standard error of the mean (SEM) Fig-ure 8 shows the mean overall time and distance requiredto reach the goal kiosk and gure 9 breaks the overalltime into its components Figure 10 gives the forwardand backtracking distance components Figure 11 showsthe mean number of times that each type of guidebook

Figure 8 (left) Mean total time The overall time spent traveling to

the goal was signicantly less using the worldlet guidebook than when

using the text (F(1 24) 5 989 p 001) or image (F(1 24) 5

463 p 001) guidebooks (right) Mean total distance The overall

distance traveled using the worldlet guidebook was signicantly less

than the distance traveled using the text (F(1 24) 5 1141 p

001) or image (F(1 24) 5 893 p 001) guidebooks

576 PRESENCE VOLUME 10 NUMBER 6

was opened Figure 12 diagrams the subjectsrsquo scores onthe Wechsler neuropsychological tests

Analysis of variance followed by Scheffe post hoc testswere performed for the different measures Table 2summarizes these results

72 Street Plots

Figure 13 shows plots of the streets traversed by the26 subjects in the three test cities The thickness andbrightness of each street line indicate the total number oftimes that the street was traversed For example widerbrighter streets indicate heavily used avenues Backtrackingcan increase the number of street traversals beyond theexpected minimum of 26 The starting point and goalkiosk are indicated on each plot and landmarks enroute tothe goal are marked with a star (q )

Figure 10 The mean distance moving forward while using the

worldlet guidebook is signicantly less than the forward distance

covered while using the text or image guidebook Backtracking

distance is almost nonexistent for the worldlet guidebook task

Figure 9 Turning looking for landmark is time spent standing in

one location Turning changing forward direction is time spent

changing forward direction Consulting guidebook is time spent with

the guidebook on screen Moving forward is time spent traveling over

new ground and Backtracking is time spent retreating Note that the

time spent consulting the worldlet guidebook is signicantly greater

than the time spent consulting the text or image guidebook Also note

that the time spent backtracking is near zero when using the worldlet

guidebook

Figure 11 Subjects opened the text and image guidebooks an

average of fteen to sixteen times while seeking the eight landmarks

enroute to the goal kiosk The worldlet guidebook was opened eleven

times on average

Figure 12 Z-transformed spatial and verbal scores from the

neuropsychological tests The mean raw scores were 45 and 55 for

the spatial and verbal tests respectively

Elvins et al 577

73 Subjective Measures

As part of the written questionnaire administeredafter the experiment subjects were asked how difcultthe search task was using each of the different types ofguidebooks Although responses were scattered a cleartrend in favor of the worldlet guidebook was evident asshown in table 3

74 Correlation Analysis

One surprising nding was a high inverse corre-lation ( 2 077 p 5 001) between spatial skills andthe time spent consulting the text guidebook Sub-jects with better spatial ability seemed able to form amental picture of a landmark based on textual de-scription more quickly than did subjects with poorer

spatial ability This is not the case for image andworldlet guidebooks however

We expected to nd a negative correlation betweenspatial skills and overall travel time in the image andworldlet guidebook cities The data did not supportthis nor did they support another of our hypothesesthat there would be a positive correlation between ver-bal skills and the performance in the text guidebookcity Verbal skills made no difference in overall waynd-ing time

75 Observations on Subject Strategies

Many subjects said that the tasks were harderthan they had imagined and afterwards several saidthat they felt tired No subjects complained of mo-

Table 2 Our ANOVA Indicated Signicant Differences between All the Sample Means and Post Hoc Analysis Was Warranted

ANOVA

Post hoc comparisons (Scheffe F-test)

text vs image text vs worldlet image vs worldlet

TimeConsulting guidebook 5559 p 00001 1966 p 0001 884 p 001 5488 p 0011Looking for landmark 2161 p 00001 047 1869 p 0001 1325 p 001Changing heading 1736 p 00001 002 1353 p 001 1249 p 001Moving forward 1602 p 00001 0003 1220 p 001 1183 p 001Backtracking 955 p 00003 022 826 p 001 586 p 001Overall 1034 p 00002 098 989 p 001 463 p 001

DistanceMoving forward 1533 p 00001 005 1223 p 001 1072 p 001Backtracking 1019 p 00002 023 884 p 001 621 p 001Overall 1365 p 00001 015 1141 p 001 893 p 001

GuidebookOpened 1527 p 00001 104 759 p 001 1426 p 0001

Post hoc (Scheffe) analysis indicated for all dependent variables signicant differences in the sample means of the textversus worldlet performance and image versus worldlet performance Performance using the text versus imageguidebooks was not signicantly different except that the time spent consulting the text guidebook was signicantlygreater than the time spent consulting the image guidebook In each table cell the rst number is the F value and thesecond number is the p value Asterisk indicates signicance p 005 Time and distance measures are explained in thecaptions for gures 8 through 11

578 PRESENCE VOLUME 10 NUMBER 6

tion sickness Most subjects required 70 min or moreto nish all aspects of the computer-based part of theexperiment

We observed subjects to use three general types ofstrategies while deciding what direction to go from astreet intersection

c Look down each street from many viewpointswithin the intersection until a landmark or land-mark context is seen

c Look down each street and if nothing looks famil-iar search exhaustively in each direction until thenext landmark is found

c Look down each street and if nothing looks famil-iar try going one-half block in each direction andlook again

Subjects using the rst strategy seemed to nd theirway more quickly and with less backtracking than didsubjects who used the second and third strategies An-other strategy described by some subjects on their writ-ten questionnaire involved viewing a landmark in theworldlet guidebook from many distant viewpoints in thehopes that one of the viewpoints would be similar to thedirection from which they would actually approach the

Figure 13 In these street plots line thickness and brightness

indicates the number of times that a street was traversed In top and

middle the 26 subjects traversed some streets 47 times indicating

backtracking Top and middle also exhibit circuitous trails In bottom

very little backtracking occurred and circuitous trails are minimal

Table 3 Subjects Were Asked to Rate the Difculty of theSearch Task Using the Three Different Types of Guidebooks

Text Image Worldlet

Very easy 0 2 11Easy 10 8 9Doable 10 8 6Difcult 5 7 0Very difcult 1 1 0

Median Doable Doable Easy

Elvins et al 579

landmark in the city We observed that some subjectsoften forgot what it was they were looking for and sohad to open the guidebook momentarily to remindthemselves

Not surprisingly landmarks with prominent signage(such as ldquoBobrsquos Groceryrdquo) and landmarks painted withmore highly saturated colors seemed easier to nd thanother types of landmarks Also after passing all thelandmarks several subjects exclaimed that they had lostall sense of direction within the world

A nal unexpected result were the many similarities be-tween the text and image guidebooks The text and imageguidebooks were only signicantly different in the timesubjects spent consulting them Considering that text andimages represent landmarks in very different ways we ex-pected differences in time and distance measures

76 Major Findings

In interpreting these ndings it is important toremember that the subjects performed the wayndingtasks using only the landmark knowledge gained bystudying the guidebooks No procedural or survey(map) information was provided

c As we predicted the total time spent traveling tothe goal was signicantly less using the worldletguidebook than when using the text or imageguidebooks

c Subjects spent signicantly more time studying theworldlet guidebook than they did the text or imageguidebook

c Increased worldlet study time was more than com-pensated for by a reduction in the time required toreach the goal Using the worldlet guidebook sub-jects traveled an average of 55 faster than whenusing the text guidebook and 38 faster than whenusing the image guidebook

c Increased worldlet study time also allowed subjectsto move more directly to the goal and reducedbacktracking time to almost zero This is showndiagramatically in gure 13 in which gures 13aand 13b show brighter wider streets than in gure

13c Figure 13c also exhibits fewer detours and lesswandering than do gures 13a and 13b

We postulate that compared to the landmark knowl-edge gained from text and image guidebooks the im-proved landmark knowledge built while consulting theworldlet guidebook enabled subjects to refer less fre-quently to the guidebook This postulation is furthersupported by examining worldlet guidebook results

c Subjects spent signicantly more time studying theworldlet guidebook than they did the other types ofguidebooks

c When using the worldlet guidebook subjects trav-eled faster more directly and with more skill tothe goal kiosk

c Subjectsrsquo median rating for the difculty of thesearch task was Easy when using the worldlet guide-book

8 Conclusion

The waynding literature shows that landmarksare critical in navigating an environment whether it isreal or virtual While building landmark knowledgetravelers characterize a landmark representation by its3-D shape size texture context and so forth In awaynding task textual or image guidebook landmarkdescriptions fail to engage the full range of 3-D land-mark characteristics recognized and used by travelers tond their way

We know of no studies that have considered the com-parative advantages of different types of guidebooks forwaynding Intuitively one would predict that 3-Dthumbnails make better guidebooks for 3-D worldsthan do text and 2-D thumbnail images and indeedthis is what we found We wanted however to identifythe component activities in using guidebooks that mostbenetted from 3-D elements Perhaps of greatest inter-est was our nding that worldlets enabled subjects tond their way in complex environments with almost nobacktracking By spending more time studying theworldlets in their guidebooks they spent less time look-ing around reorienting themselves conducting exhaus-

580 PRESENCE VOLUME 10 NUMBER 6

tive searches and performing similar time-wasting activ-ities We must be cautious however in what weconclude because worldlets contain more informationthan do text or 2-D thumbnail images To control forthis difference in information content more experi-ments are necessary

9 Future Work

We would like to investigate the comparative ad-vantages of worldlet guidebooks for waynding tasks incorresponding virtual and real world environments(both urban and rural) We are also considering an ex-periment to evaluate tandem use of traditional 2-Dmaps and landmark guidebooks Requiring effectivemap and compass use in a natural environment thesport of orienteering (Kjellstrom 1976) might benetfrom worldlets By encapsulating the region around theagged control points into a worldlet orienteers couldgain rst-person experience with natural landmarks suchas foliage rocks gullies shadows and so on beforebeginning the search

Acknowledgments

Our appreciation goes to Planet 9 Studios (wwwplanet9com)for the use of their city building textures Thanks to JohnMoreland and Mike Bailey for technical assistance on thisproject and thanks to Suzi Feeney for collaborating in thepilot study design UCSD librarian Debbie Cox and her staffwere invaluable in locating references Thanks also to JohnHelly and Reagan Moore for their support San Diego Super-computer Center is funded by NSF the University of Califor-nia the State of California and industrial partners

References

Appleyard D A (1976) Planning a Pluralistic City Con-icting Realities in Ciudad Guayana Cambridge MAMIT Press

Art1 Com (1997) The cybercity Berlin project [on-line]Available httpwwwartcomdeprojects

Chen S E (1995) QuickTime VRmdashAn image-based ap-proach to VE navigation Proceedings of SIGGRAPH 9529 ndash38

Darken R P amp Sibert J L (1996) Waynding strategiesand behaviors in large virtual worlds Proceedings of CHI 96142ndash149

Downs R J amp Stea D (1973) Cognitive maps and spatialbehavior In R J Downs amp D Stea (Eds) Image and En-vironment Chicago Aldine Publishing Company 8 ndash26

Elvins T T (1997) Waynding in layered informationworlds Proceedings of the CODATA Euro-American Work-shop on Data and Information Visualization 6 ndash7

Elvins T T Nadeau D R amp Kirsh D (1997) World-letsmdash3D thumbnails for waynding in virtual environmentsProceedings of the ACM User Interface Software and Technol-ogy Symposium 21ndash30

Elvins T T (1998) WorldletsmdashA novel 3D thumbnail af-fordance for waynding in large virtual environmentsUnpublished doctoral dissertation Department of Elec-trical and Computer Engineering University of Califor-niandashSan Diego

Elvins T T Nadeau D R Kirsh D amp Schul R (1998)Worldletsmdash3D thumbnails for 3D browsing Proceedings ofthe ACM CHI Conference 21ndash30

Foley J van Dam A Feiner S amp Hughes J (1990) Com-puter Graphics Principles and Practice Reading MA Addi-son-Wesley

Gillner S amp Mallot H A (1998) Navigation and acquisi-tion of spatial knowledge in a virtual maze Journal of Cog-nitive Neuroscience 10(4) 445ndash 463

Goldin S E amp Thorndyke P W (1982) Simulating naviga-tion for spatial knowledge acquisition Human Factors24(4) 457ndash 471

ISO (1997) The Virtual Reality Modeling Language Inter-national Standard ISOIEC 14772-11997 [on-line] Avail-able httpwwwvrmlorgSpecicationsVRML97

Kjellstrom B (1976) Be Expert with Map and Compass TheComplete Orienteering Handbook New York Scribner Pub-lishers

Liggett R Friedman S amp Jepson W (1995) Interactivedesigndecision making in a virtual urban world visual sim-ulation and GIS Proceedings of the 1995 ESRI User Confer-ence Environmental Systems Research [on-line] Availablehttpwwwesricombasecommonuserconfproc95

Lynch K (1960) The Image of the City Cambridge MAMIT Press

Elvins et al 581

street grid with pavement roads and sidewalks betweenthe blocks Each block contained up to twenty build-ings side by side around the blockrsquos perimeter Using acache of 250 building designs buildings were randomlyselected and placed on city blocks Many non-landmarkbuildings were repeated between and within the citiesRendering requirements were reduced by programmingbuildings to be invisible when farther than three blocksfrom the current viewpoint To prevent subjects fromseeing buildings appearing and disappearing in the dis-tance we added fog to the environment

Cities contained parks parking lots and other civicfeatures but did not contain cars or street signs Land-marks were placed so that a subject could not short-cutthe route and would never have to make an exhaustivesearch as long as they carefully looked down each possi-ble path for familiar landmarks and landmark contextthey could always determine in which direction to pro-ceed No landmarks were located within one block ofthe perimeter wall that surrounded each city

64 Radius of Information

The radius of information was the most difcultparameter to make constant across the three guide-

books Given that text 2-D images and worldlets aredifferent representations they do not convey an equiva-lent quantity and quality of landmark information Wedetermined while building the guidebooks for the pilotstudy that using a radius of 20 m produced the mostnearly equivalent set of guidebooks For the text guide-book this meant describing in words the landmark thebuilding or structure directly across the street thebuilding on each side and whether the landmark was ona corner Thus our textual description of the landmarkshown in gure 3 read

Citgo Gas Station Citgorsquos red gas pumps are at acorner location Next door is the Cafe 360 in a beigecolored building and across the street is the LumbiniRestaurant in a red brick building

For the image guidebook a 20 m radius meant posi-tioning the camera across the street from the landmarkso that the captured thumbnail image would contain thebuildings or structures on the left and right of the land-mark For the worldlet guidebook setting the informa-tion radius simply meant setting the yon clip plane tocutaway 3-D shapes farther away than 20 m while cap-turing a spherical worldlet Image and worldlet camera

Table 1 Although the Arrangement of Buildings Landmarks and Routes Were Different in Each City Their Basic Structurein Terms of City Design and Landmark Layout was Kept Constant to Allow for Controlled Comparison Subjects Movedat a Fixed Speed in all Cities

Parameter Constant value Comments

City dimensions ten by ten blocks 100 total blocksBuilding lots per block twenty on perimeter of block building footprint variesDistance from start to goal 1200 meters past 100 12-meter-wide lotsLandmarks per city nine including goal kiosk ordered in guidebooksDistance between landmarks 15 blocks 6 025 blockTravel speed ve meterssecond slow driving speedTypes of landmarks eight one of each kind in each cityTurns required to reach goal ve left and right turnsRadius of information approximately twenty meters see city design sectionLandmark positions three on corner ve midblock goal is midblockGoal building yellowred striped kiosk six-sided viewpoint invariant

574 PRESENCE VOLUME 10 NUMBER 6

locations were selected so as not to suggest a directionof approach to the landmark

65 Hardware and Software Design

The experiment was conducted using a standard19 in monitor Stereo glasses were not used A maincity window displayed the city in the waynding VRMLbrowser Subjects used keyboard arrow keys to moveforward and backward a xed distance on each keypress and to turn left and right by a xed angle on eachkey press In the worldlet guidebook subjects used thearrow keys in the same manner to move around in theworldlet Collision detection prevented subjects frompassing through buildings and other objects in the city

Subjects pushed a Start button before beginning aLandmark button each time a landmark was reachedand a Stop button upon reaching the goal kiosk Sub-jects pushed a Guidebook button to display the guide-book Subjects practiced operating all of these buttonsin practice cities prior to beginning the experiment Thewindows were designed so that when the guidebookwas open (shown in gure 4) the main city window(shown in gure 7) was completely occluded and nomovement could be made

66 Lessons Learned from the PilotStudy

Our pilot study design and results are described byElvins et al (1997) From the pilot study we learnedthat landmarks cannot be more than two blocks apartbecause subjects frequently resort to an exhaustivesearch of the neighborhood We also learned that somesubjects will take much longer than other subjects tocomplete the task and that the task difculty and lengthmust be designed while keeping the slower wayndersin mind

67 Subject Instructions

The following is a subset of the instructions readto the subjects Instructions 3 and 4 were repeated be-fore beginning each city

1 The guidebook contains a series of landmarks thatyou probably want to follow since traveling pastthese landmarks will lead you to the goal kiosk

2 When you are at a landmark you may have towalk as far as one or two blocks before you will beable to see the next landmark

Figure 7 The main city window

Elvins et al 575

3 Your task is to locate the goal kiosk as quickly aspossible Go as fast as you can but be careful tonot get lost

4 Remember you can look at the guidebook as of-ten as you like

68 Subjects

Subjects were primarily University of CaliforniandashSan Diego students from a number of major depart-ments Ages ranged from 18 to 36 with a mean age of24 Fourteen men and twelve women completed theexperiment Results from four other subjects had to beexcluded from our data because the subjects becamelost in one or more of the cities All subjects were com-puter literate but with varying degrees of experiencewith VEs

69 Neuropsychological Tests

To provide an independent measure of a subset ofthe subjectsrsquo cognitive abilities subjects were tested in aseparate session using two standardized subtests fromthe Wechsler Adult Intelligent Scale-Revised (WAIS-R)(Wechsler 1955) The Vocabulary subtest comprisingpart of the WAIS-R Verbal subtests was used to assessthe subjectrsquos verbal ability Performance was scored ac-cording to standardized procedures based on the sub-jectrsquos degree of familiarity with a list of up to 35 wordsand their elegance of expression The second test wasthe Block Design subtest comprising part of theWAIS-R Performance subtests which was used to assessthe subjectsrsquo constructional-spatial ability and was ad-ministered in the standardized manner

610 Analysis

A one-way repeated measures analysis of variance(ANOVA) was performed for each of the dependentvariables and the overall times and distances The with-in-subjects variable was the landmark description typewith three levels text image and worldlet Post hocanalyses were completed using the Scheffe test for dif-ferences in sample means Correlation analysis was per-

formed across the scores of the neuropsychological testsand the scores of the waynding tests We adopted asignicance level of 005 on all analyses unless otherwisenoted

7 Results

Our primary result was that people spend moretime consulting the worldlet guidebook than the otherguidebooks but that they require less time to reach thegoal when using the worldlet guidebook The followingspecic ndings tell us more about the component ac-tivities that benetted from worldlet use

71 Time and Distance Measures

Figures 8 through 11 show time and distancemeasures in which the independent variable (type ofguidebook) is displayed as white gray and black barsalong with the standard error of the mean (SEM) Fig-ure 8 shows the mean overall time and distance requiredto reach the goal kiosk and gure 9 breaks the overalltime into its components Figure 10 gives the forwardand backtracking distance components Figure 11 showsthe mean number of times that each type of guidebook

Figure 8 (left) Mean total time The overall time spent traveling to

the goal was signicantly less using the worldlet guidebook than when

using the text (F(1 24) 5 989 p 001) or image (F(1 24) 5

463 p 001) guidebooks (right) Mean total distance The overall

distance traveled using the worldlet guidebook was signicantly less

than the distance traveled using the text (F(1 24) 5 1141 p

001) or image (F(1 24) 5 893 p 001) guidebooks

576 PRESENCE VOLUME 10 NUMBER 6

was opened Figure 12 diagrams the subjectsrsquo scores onthe Wechsler neuropsychological tests

Analysis of variance followed by Scheffe post hoc testswere performed for the different measures Table 2summarizes these results

72 Street Plots

Figure 13 shows plots of the streets traversed by the26 subjects in the three test cities The thickness andbrightness of each street line indicate the total number oftimes that the street was traversed For example widerbrighter streets indicate heavily used avenues Backtrackingcan increase the number of street traversals beyond theexpected minimum of 26 The starting point and goalkiosk are indicated on each plot and landmarks enroute tothe goal are marked with a star (q )

Figure 10 The mean distance moving forward while using the

worldlet guidebook is signicantly less than the forward distance

covered while using the text or image guidebook Backtracking

distance is almost nonexistent for the worldlet guidebook task

Figure 9 Turning looking for landmark is time spent standing in

one location Turning changing forward direction is time spent

changing forward direction Consulting guidebook is time spent with

the guidebook on screen Moving forward is time spent traveling over

new ground and Backtracking is time spent retreating Note that the

time spent consulting the worldlet guidebook is signicantly greater

than the time spent consulting the text or image guidebook Also note

that the time spent backtracking is near zero when using the worldlet

guidebook

Figure 11 Subjects opened the text and image guidebooks an

average of fteen to sixteen times while seeking the eight landmarks

enroute to the goal kiosk The worldlet guidebook was opened eleven

times on average

Figure 12 Z-transformed spatial and verbal scores from the

neuropsychological tests The mean raw scores were 45 and 55 for

the spatial and verbal tests respectively

Elvins et al 577

73 Subjective Measures

As part of the written questionnaire administeredafter the experiment subjects were asked how difcultthe search task was using each of the different types ofguidebooks Although responses were scattered a cleartrend in favor of the worldlet guidebook was evident asshown in table 3

74 Correlation Analysis

One surprising nding was a high inverse corre-lation ( 2 077 p 5 001) between spatial skills andthe time spent consulting the text guidebook Sub-jects with better spatial ability seemed able to form amental picture of a landmark based on textual de-scription more quickly than did subjects with poorer

spatial ability This is not the case for image andworldlet guidebooks however

We expected to nd a negative correlation betweenspatial skills and overall travel time in the image andworldlet guidebook cities The data did not supportthis nor did they support another of our hypothesesthat there would be a positive correlation between ver-bal skills and the performance in the text guidebookcity Verbal skills made no difference in overall waynd-ing time

75 Observations on Subject Strategies

Many subjects said that the tasks were harderthan they had imagined and afterwards several saidthat they felt tired No subjects complained of mo-

Table 2 Our ANOVA Indicated Signicant Differences between All the Sample Means and Post Hoc Analysis Was Warranted

ANOVA

Post hoc comparisons (Scheffe F-test)

text vs image text vs worldlet image vs worldlet

TimeConsulting guidebook 5559 p 00001 1966 p 0001 884 p 001 5488 p 0011Looking for landmark 2161 p 00001 047 1869 p 0001 1325 p 001Changing heading 1736 p 00001 002 1353 p 001 1249 p 001Moving forward 1602 p 00001 0003 1220 p 001 1183 p 001Backtracking 955 p 00003 022 826 p 001 586 p 001Overall 1034 p 00002 098 989 p 001 463 p 001

DistanceMoving forward 1533 p 00001 005 1223 p 001 1072 p 001Backtracking 1019 p 00002 023 884 p 001 621 p 001Overall 1365 p 00001 015 1141 p 001 893 p 001

GuidebookOpened 1527 p 00001 104 759 p 001 1426 p 0001

Post hoc (Scheffe) analysis indicated for all dependent variables signicant differences in the sample means of the textversus worldlet performance and image versus worldlet performance Performance using the text versus imageguidebooks was not signicantly different except that the time spent consulting the text guidebook was signicantlygreater than the time spent consulting the image guidebook In each table cell the rst number is the F value and thesecond number is the p value Asterisk indicates signicance p 005 Time and distance measures are explained in thecaptions for gures 8 through 11

578 PRESENCE VOLUME 10 NUMBER 6

tion sickness Most subjects required 70 min or moreto nish all aspects of the computer-based part of theexperiment

We observed subjects to use three general types ofstrategies while deciding what direction to go from astreet intersection

c Look down each street from many viewpointswithin the intersection until a landmark or land-mark context is seen

c Look down each street and if nothing looks famil-iar search exhaustively in each direction until thenext landmark is found

c Look down each street and if nothing looks famil-iar try going one-half block in each direction andlook again

Subjects using the rst strategy seemed to nd theirway more quickly and with less backtracking than didsubjects who used the second and third strategies An-other strategy described by some subjects on their writ-ten questionnaire involved viewing a landmark in theworldlet guidebook from many distant viewpoints in thehopes that one of the viewpoints would be similar to thedirection from which they would actually approach the

Figure 13 In these street plots line thickness and brightness

indicates the number of times that a street was traversed In top and

middle the 26 subjects traversed some streets 47 times indicating

backtracking Top and middle also exhibit circuitous trails In bottom

very little backtracking occurred and circuitous trails are minimal

Table 3 Subjects Were Asked to Rate the Difculty of theSearch Task Using the Three Different Types of Guidebooks

Text Image Worldlet

Very easy 0 2 11Easy 10 8 9Doable 10 8 6Difcult 5 7 0Very difcult 1 1 0

Median Doable Doable Easy

Elvins et al 579

landmark in the city We observed that some subjectsoften forgot what it was they were looking for and sohad to open the guidebook momentarily to remindthemselves

Not surprisingly landmarks with prominent signage(such as ldquoBobrsquos Groceryrdquo) and landmarks painted withmore highly saturated colors seemed easier to nd thanother types of landmarks Also after passing all thelandmarks several subjects exclaimed that they had lostall sense of direction within the world

A nal unexpected result were the many similarities be-tween the text and image guidebooks The text and imageguidebooks were only signicantly different in the timesubjects spent consulting them Considering that text andimages represent landmarks in very different ways we ex-pected differences in time and distance measures

76 Major Findings

In interpreting these ndings it is important toremember that the subjects performed the wayndingtasks using only the landmark knowledge gained bystudying the guidebooks No procedural or survey(map) information was provided

c As we predicted the total time spent traveling tothe goal was signicantly less using the worldletguidebook than when using the text or imageguidebooks

c Subjects spent signicantly more time studying theworldlet guidebook than they did the text or imageguidebook

c Increased worldlet study time was more than com-pensated for by a reduction in the time required toreach the goal Using the worldlet guidebook sub-jects traveled an average of 55 faster than whenusing the text guidebook and 38 faster than whenusing the image guidebook

c Increased worldlet study time also allowed subjectsto move more directly to the goal and reducedbacktracking time to almost zero This is showndiagramatically in gure 13 in which gures 13aand 13b show brighter wider streets than in gure

13c Figure 13c also exhibits fewer detours and lesswandering than do gures 13a and 13b

We postulate that compared to the landmark knowl-edge gained from text and image guidebooks the im-proved landmark knowledge built while consulting theworldlet guidebook enabled subjects to refer less fre-quently to the guidebook This postulation is furthersupported by examining worldlet guidebook results

c Subjects spent signicantly more time studying theworldlet guidebook than they did the other types ofguidebooks

c When using the worldlet guidebook subjects trav-eled faster more directly and with more skill tothe goal kiosk

c Subjectsrsquo median rating for the difculty of thesearch task was Easy when using the worldlet guide-book

8 Conclusion

The waynding literature shows that landmarksare critical in navigating an environment whether it isreal or virtual While building landmark knowledgetravelers characterize a landmark representation by its3-D shape size texture context and so forth In awaynding task textual or image guidebook landmarkdescriptions fail to engage the full range of 3-D land-mark characteristics recognized and used by travelers tond their way

We know of no studies that have considered the com-parative advantages of different types of guidebooks forwaynding Intuitively one would predict that 3-Dthumbnails make better guidebooks for 3-D worldsthan do text and 2-D thumbnail images and indeedthis is what we found We wanted however to identifythe component activities in using guidebooks that mostbenetted from 3-D elements Perhaps of greatest inter-est was our nding that worldlets enabled subjects tond their way in complex environments with almost nobacktracking By spending more time studying theworldlets in their guidebooks they spent less time look-ing around reorienting themselves conducting exhaus-

580 PRESENCE VOLUME 10 NUMBER 6

tive searches and performing similar time-wasting activ-ities We must be cautious however in what weconclude because worldlets contain more informationthan do text or 2-D thumbnail images To control forthis difference in information content more experi-ments are necessary

9 Future Work

We would like to investigate the comparative ad-vantages of worldlet guidebooks for waynding tasks incorresponding virtual and real world environments(both urban and rural) We are also considering an ex-periment to evaluate tandem use of traditional 2-Dmaps and landmark guidebooks Requiring effectivemap and compass use in a natural environment thesport of orienteering (Kjellstrom 1976) might benetfrom worldlets By encapsulating the region around theagged control points into a worldlet orienteers couldgain rst-person experience with natural landmarks suchas foliage rocks gullies shadows and so on beforebeginning the search

Acknowledgments

Our appreciation goes to Planet 9 Studios (wwwplanet9com)for the use of their city building textures Thanks to JohnMoreland and Mike Bailey for technical assistance on thisproject and thanks to Suzi Feeney for collaborating in thepilot study design UCSD librarian Debbie Cox and her staffwere invaluable in locating references Thanks also to JohnHelly and Reagan Moore for their support San Diego Super-computer Center is funded by NSF the University of Califor-nia the State of California and industrial partners

References

Appleyard D A (1976) Planning a Pluralistic City Con-icting Realities in Ciudad Guayana Cambridge MAMIT Press

Art1 Com (1997) The cybercity Berlin project [on-line]Available httpwwwartcomdeprojects

Chen S E (1995) QuickTime VRmdashAn image-based ap-proach to VE navigation Proceedings of SIGGRAPH 9529 ndash38

Darken R P amp Sibert J L (1996) Waynding strategiesand behaviors in large virtual worlds Proceedings of CHI 96142ndash149

Downs R J amp Stea D (1973) Cognitive maps and spatialbehavior In R J Downs amp D Stea (Eds) Image and En-vironment Chicago Aldine Publishing Company 8 ndash26

Elvins T T (1997) Waynding in layered informationworlds Proceedings of the CODATA Euro-American Work-shop on Data and Information Visualization 6 ndash7

Elvins T T Nadeau D R amp Kirsh D (1997) World-letsmdash3D thumbnails for waynding in virtual environmentsProceedings of the ACM User Interface Software and Technol-ogy Symposium 21ndash30

Elvins T T (1998) WorldletsmdashA novel 3D thumbnail af-fordance for waynding in large virtual environmentsUnpublished doctoral dissertation Department of Elec-trical and Computer Engineering University of Califor-niandashSan Diego

Elvins T T Nadeau D R Kirsh D amp Schul R (1998)Worldletsmdash3D thumbnails for 3D browsing Proceedings ofthe ACM CHI Conference 21ndash30

Foley J van Dam A Feiner S amp Hughes J (1990) Com-puter Graphics Principles and Practice Reading MA Addi-son-Wesley

Gillner S amp Mallot H A (1998) Navigation and acquisi-tion of spatial knowledge in a virtual maze Journal of Cog-nitive Neuroscience 10(4) 445ndash 463

Goldin S E amp Thorndyke P W (1982) Simulating naviga-tion for spatial knowledge acquisition Human Factors24(4) 457ndash 471

ISO (1997) The Virtual Reality Modeling Language Inter-national Standard ISOIEC 14772-11997 [on-line] Avail-able httpwwwvrmlorgSpecicationsVRML97

Kjellstrom B (1976) Be Expert with Map and Compass TheComplete Orienteering Handbook New York Scribner Pub-lishers

Liggett R Friedman S amp Jepson W (1995) Interactivedesigndecision making in a virtual urban world visual sim-ulation and GIS Proceedings of the 1995 ESRI User Confer-ence Environmental Systems Research [on-line] Availablehttpwwwesricombasecommonuserconfproc95

Lynch K (1960) The Image of the City Cambridge MAMIT Press

Elvins et al 581

locations were selected so as not to suggest a directionof approach to the landmark

65 Hardware and Software Design

The experiment was conducted using a standard19 in monitor Stereo glasses were not used A maincity window displayed the city in the waynding VRMLbrowser Subjects used keyboard arrow keys to moveforward and backward a xed distance on each keypress and to turn left and right by a xed angle on eachkey press In the worldlet guidebook subjects used thearrow keys in the same manner to move around in theworldlet Collision detection prevented subjects frompassing through buildings and other objects in the city

Subjects pushed a Start button before beginning aLandmark button each time a landmark was reachedand a Stop button upon reaching the goal kiosk Sub-jects pushed a Guidebook button to display the guide-book Subjects practiced operating all of these buttonsin practice cities prior to beginning the experiment Thewindows were designed so that when the guidebookwas open (shown in gure 4) the main city window(shown in gure 7) was completely occluded and nomovement could be made

66 Lessons Learned from the PilotStudy

Our pilot study design and results are described byElvins et al (1997) From the pilot study we learnedthat landmarks cannot be more than two blocks apartbecause subjects frequently resort to an exhaustivesearch of the neighborhood We also learned that somesubjects will take much longer than other subjects tocomplete the task and that the task difculty and lengthmust be designed while keeping the slower wayndersin mind

67 Subject Instructions

The following is a subset of the instructions readto the subjects Instructions 3 and 4 were repeated be-fore beginning each city

1 The guidebook contains a series of landmarks thatyou probably want to follow since traveling pastthese landmarks will lead you to the goal kiosk

2 When you are at a landmark you may have towalk as far as one or two blocks before you will beable to see the next landmark

Figure 7 The main city window

Elvins et al 575

3 Your task is to locate the goal kiosk as quickly aspossible Go as fast as you can but be careful tonot get lost

4 Remember you can look at the guidebook as of-ten as you like

68 Subjects

Subjects were primarily University of CaliforniandashSan Diego students from a number of major depart-ments Ages ranged from 18 to 36 with a mean age of24 Fourteen men and twelve women completed theexperiment Results from four other subjects had to beexcluded from our data because the subjects becamelost in one or more of the cities All subjects were com-puter literate but with varying degrees of experiencewith VEs

69 Neuropsychological Tests

To provide an independent measure of a subset ofthe subjectsrsquo cognitive abilities subjects were tested in aseparate session using two standardized subtests fromthe Wechsler Adult Intelligent Scale-Revised (WAIS-R)(Wechsler 1955) The Vocabulary subtest comprisingpart of the WAIS-R Verbal subtests was used to assessthe subjectrsquos verbal ability Performance was scored ac-cording to standardized procedures based on the sub-jectrsquos degree of familiarity with a list of up to 35 wordsand their elegance of expression The second test wasthe Block Design subtest comprising part of theWAIS-R Performance subtests which was used to assessthe subjectsrsquo constructional-spatial ability and was ad-ministered in the standardized manner

610 Analysis

A one-way repeated measures analysis of variance(ANOVA) was performed for each of the dependentvariables and the overall times and distances The with-in-subjects variable was the landmark description typewith three levels text image and worldlet Post hocanalyses were completed using the Scheffe test for dif-ferences in sample means Correlation analysis was per-

formed across the scores of the neuropsychological testsand the scores of the waynding tests We adopted asignicance level of 005 on all analyses unless otherwisenoted

7 Results

Our primary result was that people spend moretime consulting the worldlet guidebook than the otherguidebooks but that they require less time to reach thegoal when using the worldlet guidebook The followingspecic ndings tell us more about the component ac-tivities that benetted from worldlet use

71 Time and Distance Measures

Figures 8 through 11 show time and distancemeasures in which the independent variable (type ofguidebook) is displayed as white gray and black barsalong with the standard error of the mean (SEM) Fig-ure 8 shows the mean overall time and distance requiredto reach the goal kiosk and gure 9 breaks the overalltime into its components Figure 10 gives the forwardand backtracking distance components Figure 11 showsthe mean number of times that each type of guidebook

Figure 8 (left) Mean total time The overall time spent traveling to

the goal was signicantly less using the worldlet guidebook than when

using the text (F(1 24) 5 989 p 001) or image (F(1 24) 5

463 p 001) guidebooks (right) Mean total distance The overall

distance traveled using the worldlet guidebook was signicantly less

than the distance traveled using the text (F(1 24) 5 1141 p

001) or image (F(1 24) 5 893 p 001) guidebooks

576 PRESENCE VOLUME 10 NUMBER 6

was opened Figure 12 diagrams the subjectsrsquo scores onthe Wechsler neuropsychological tests

Analysis of variance followed by Scheffe post hoc testswere performed for the different measures Table 2summarizes these results

72 Street Plots

Figure 13 shows plots of the streets traversed by the26 subjects in the three test cities The thickness andbrightness of each street line indicate the total number oftimes that the street was traversed For example widerbrighter streets indicate heavily used avenues Backtrackingcan increase the number of street traversals beyond theexpected minimum of 26 The starting point and goalkiosk are indicated on each plot and landmarks enroute tothe goal are marked with a star (q )

Figure 10 The mean distance moving forward while using the

worldlet guidebook is signicantly less than the forward distance

covered while using the text or image guidebook Backtracking

distance is almost nonexistent for the worldlet guidebook task

Figure 9 Turning looking for landmark is time spent standing in

one location Turning changing forward direction is time spent

changing forward direction Consulting guidebook is time spent with

the guidebook on screen Moving forward is time spent traveling over

new ground and Backtracking is time spent retreating Note that the

time spent consulting the worldlet guidebook is signicantly greater

than the time spent consulting the text or image guidebook Also note

that the time spent backtracking is near zero when using the worldlet

guidebook

Figure 11 Subjects opened the text and image guidebooks an

average of fteen to sixteen times while seeking the eight landmarks

enroute to the goal kiosk The worldlet guidebook was opened eleven

times on average

Figure 12 Z-transformed spatial and verbal scores from the

neuropsychological tests The mean raw scores were 45 and 55 for

the spatial and verbal tests respectively

Elvins et al 577

73 Subjective Measures

As part of the written questionnaire administeredafter the experiment subjects were asked how difcultthe search task was using each of the different types ofguidebooks Although responses were scattered a cleartrend in favor of the worldlet guidebook was evident asshown in table 3

74 Correlation Analysis

One surprising nding was a high inverse corre-lation ( 2 077 p 5 001) between spatial skills andthe time spent consulting the text guidebook Sub-jects with better spatial ability seemed able to form amental picture of a landmark based on textual de-scription more quickly than did subjects with poorer

spatial ability This is not the case for image andworldlet guidebooks however

We expected to nd a negative correlation betweenspatial skills and overall travel time in the image andworldlet guidebook cities The data did not supportthis nor did they support another of our hypothesesthat there would be a positive correlation between ver-bal skills and the performance in the text guidebookcity Verbal skills made no difference in overall waynd-ing time

75 Observations on Subject Strategies

Many subjects said that the tasks were harderthan they had imagined and afterwards several saidthat they felt tired No subjects complained of mo-

Table 2 Our ANOVA Indicated Signicant Differences between All the Sample Means and Post Hoc Analysis Was Warranted

ANOVA

Post hoc comparisons (Scheffe F-test)

text vs image text vs worldlet image vs worldlet

TimeConsulting guidebook 5559 p 00001 1966 p 0001 884 p 001 5488 p 0011Looking for landmark 2161 p 00001 047 1869 p 0001 1325 p 001Changing heading 1736 p 00001 002 1353 p 001 1249 p 001Moving forward 1602 p 00001 0003 1220 p 001 1183 p 001Backtracking 955 p 00003 022 826 p 001 586 p 001Overall 1034 p 00002 098 989 p 001 463 p 001

DistanceMoving forward 1533 p 00001 005 1223 p 001 1072 p 001Backtracking 1019 p 00002 023 884 p 001 621 p 001Overall 1365 p 00001 015 1141 p 001 893 p 001

GuidebookOpened 1527 p 00001 104 759 p 001 1426 p 0001

Post hoc (Scheffe) analysis indicated for all dependent variables signicant differences in the sample means of the textversus worldlet performance and image versus worldlet performance Performance using the text versus imageguidebooks was not signicantly different except that the time spent consulting the text guidebook was signicantlygreater than the time spent consulting the image guidebook In each table cell the rst number is the F value and thesecond number is the p value Asterisk indicates signicance p 005 Time and distance measures are explained in thecaptions for gures 8 through 11

578 PRESENCE VOLUME 10 NUMBER 6

tion sickness Most subjects required 70 min or moreto nish all aspects of the computer-based part of theexperiment

We observed subjects to use three general types ofstrategies while deciding what direction to go from astreet intersection

c Look down each street from many viewpointswithin the intersection until a landmark or land-mark context is seen

c Look down each street and if nothing looks famil-iar search exhaustively in each direction until thenext landmark is found

c Look down each street and if nothing looks famil-iar try going one-half block in each direction andlook again

Subjects using the rst strategy seemed to nd theirway more quickly and with less backtracking than didsubjects who used the second and third strategies An-other strategy described by some subjects on their writ-ten questionnaire involved viewing a landmark in theworldlet guidebook from many distant viewpoints in thehopes that one of the viewpoints would be similar to thedirection from which they would actually approach the

Figure 13 In these street plots line thickness and brightness

indicates the number of times that a street was traversed In top and

middle the 26 subjects traversed some streets 47 times indicating

backtracking Top and middle also exhibit circuitous trails In bottom

very little backtracking occurred and circuitous trails are minimal

Table 3 Subjects Were Asked to Rate the Difculty of theSearch Task Using the Three Different Types of Guidebooks

Text Image Worldlet

Very easy 0 2 11Easy 10 8 9Doable 10 8 6Difcult 5 7 0Very difcult 1 1 0

Median Doable Doable Easy

Elvins et al 579

landmark in the city We observed that some subjectsoften forgot what it was they were looking for and sohad to open the guidebook momentarily to remindthemselves

Not surprisingly landmarks with prominent signage(such as ldquoBobrsquos Groceryrdquo) and landmarks painted withmore highly saturated colors seemed easier to nd thanother types of landmarks Also after passing all thelandmarks several subjects exclaimed that they had lostall sense of direction within the world

A nal unexpected result were the many similarities be-tween the text and image guidebooks The text and imageguidebooks were only signicantly different in the timesubjects spent consulting them Considering that text andimages represent landmarks in very different ways we ex-pected differences in time and distance measures

76 Major Findings

In interpreting these ndings it is important toremember that the subjects performed the wayndingtasks using only the landmark knowledge gained bystudying the guidebooks No procedural or survey(map) information was provided

c As we predicted the total time spent traveling tothe goal was signicantly less using the worldletguidebook than when using the text or imageguidebooks

c Subjects spent signicantly more time studying theworldlet guidebook than they did the text or imageguidebook

c Increased worldlet study time was more than com-pensated for by a reduction in the time required toreach the goal Using the worldlet guidebook sub-jects traveled an average of 55 faster than whenusing the text guidebook and 38 faster than whenusing the image guidebook

c Increased worldlet study time also allowed subjectsto move more directly to the goal and reducedbacktracking time to almost zero This is showndiagramatically in gure 13 in which gures 13aand 13b show brighter wider streets than in gure

13c Figure 13c also exhibits fewer detours and lesswandering than do gures 13a and 13b

We postulate that compared to the landmark knowl-edge gained from text and image guidebooks the im-proved landmark knowledge built while consulting theworldlet guidebook enabled subjects to refer less fre-quently to the guidebook This postulation is furthersupported by examining worldlet guidebook results

c Subjects spent signicantly more time studying theworldlet guidebook than they did the other types ofguidebooks

c When using the worldlet guidebook subjects trav-eled faster more directly and with more skill tothe goal kiosk

c Subjectsrsquo median rating for the difculty of thesearch task was Easy when using the worldlet guide-book

8 Conclusion

The waynding literature shows that landmarksare critical in navigating an environment whether it isreal or virtual While building landmark knowledgetravelers characterize a landmark representation by its3-D shape size texture context and so forth In awaynding task textual or image guidebook landmarkdescriptions fail to engage the full range of 3-D land-mark characteristics recognized and used by travelers tond their way

We know of no studies that have considered the com-parative advantages of different types of guidebooks forwaynding Intuitively one would predict that 3-Dthumbnails make better guidebooks for 3-D worldsthan do text and 2-D thumbnail images and indeedthis is what we found We wanted however to identifythe component activities in using guidebooks that mostbenetted from 3-D elements Perhaps of greatest inter-est was our nding that worldlets enabled subjects tond their way in complex environments with almost nobacktracking By spending more time studying theworldlets in their guidebooks they spent less time look-ing around reorienting themselves conducting exhaus-

580 PRESENCE VOLUME 10 NUMBER 6

tive searches and performing similar time-wasting activ-ities We must be cautious however in what weconclude because worldlets contain more informationthan do text or 2-D thumbnail images To control forthis difference in information content more experi-ments are necessary

9 Future Work

We would like to investigate the comparative ad-vantages of worldlet guidebooks for waynding tasks incorresponding virtual and real world environments(both urban and rural) We are also considering an ex-periment to evaluate tandem use of traditional 2-Dmaps and landmark guidebooks Requiring effectivemap and compass use in a natural environment thesport of orienteering (Kjellstrom 1976) might benetfrom worldlets By encapsulating the region around theagged control points into a worldlet orienteers couldgain rst-person experience with natural landmarks suchas foliage rocks gullies shadows and so on beforebeginning the search

Acknowledgments

Our appreciation goes to Planet 9 Studios (wwwplanet9com)for the use of their city building textures Thanks to JohnMoreland and Mike Bailey for technical assistance on thisproject and thanks to Suzi Feeney for collaborating in thepilot study design UCSD librarian Debbie Cox and her staffwere invaluable in locating references Thanks also to JohnHelly and Reagan Moore for their support San Diego Super-computer Center is funded by NSF the University of Califor-nia the State of California and industrial partners

References

Appleyard D A (1976) Planning a Pluralistic City Con-icting Realities in Ciudad Guayana Cambridge MAMIT Press

Art1 Com (1997) The cybercity Berlin project [on-line]Available httpwwwartcomdeprojects

Chen S E (1995) QuickTime VRmdashAn image-based ap-proach to VE navigation Proceedings of SIGGRAPH 9529 ndash38

Darken R P amp Sibert J L (1996) Waynding strategiesand behaviors in large virtual worlds Proceedings of CHI 96142ndash149

Downs R J amp Stea D (1973) Cognitive maps and spatialbehavior In R J Downs amp D Stea (Eds) Image and En-vironment Chicago Aldine Publishing Company 8 ndash26

Elvins T T (1997) Waynding in layered informationworlds Proceedings of the CODATA Euro-American Work-shop on Data and Information Visualization 6 ndash7

Elvins T T Nadeau D R amp Kirsh D (1997) World-letsmdash3D thumbnails for waynding in virtual environmentsProceedings of the ACM User Interface Software and Technol-ogy Symposium 21ndash30

Elvins T T (1998) WorldletsmdashA novel 3D thumbnail af-fordance for waynding in large virtual environmentsUnpublished doctoral dissertation Department of Elec-trical and Computer Engineering University of Califor-niandashSan Diego

Elvins T T Nadeau D R Kirsh D amp Schul R (1998)Worldletsmdash3D thumbnails for 3D browsing Proceedings ofthe ACM CHI Conference 21ndash30

Foley J van Dam A Feiner S amp Hughes J (1990) Com-puter Graphics Principles and Practice Reading MA Addi-son-Wesley

Gillner S amp Mallot H A (1998) Navigation and acquisi-tion of spatial knowledge in a virtual maze Journal of Cog-nitive Neuroscience 10(4) 445ndash 463

Goldin S E amp Thorndyke P W (1982) Simulating naviga-tion for spatial knowledge acquisition Human Factors24(4) 457ndash 471

ISO (1997) The Virtual Reality Modeling Language Inter-national Standard ISOIEC 14772-11997 [on-line] Avail-able httpwwwvrmlorgSpecicationsVRML97

Kjellstrom B (1976) Be Expert with Map and Compass TheComplete Orienteering Handbook New York Scribner Pub-lishers

Liggett R Friedman S amp Jepson W (1995) Interactivedesigndecision making in a virtual urban world visual sim-ulation and GIS Proceedings of the 1995 ESRI User Confer-ence Environmental Systems Research [on-line] Availablehttpwwwesricombasecommonuserconfproc95

Lynch K (1960) The Image of the City Cambridge MAMIT Press

Elvins et al 581

3 Your task is to locate the goal kiosk as quickly aspossible Go as fast as you can but be careful tonot get lost

4 Remember you can look at the guidebook as of-ten as you like

68 Subjects

Subjects were primarily University of CaliforniandashSan Diego students from a number of major depart-ments Ages ranged from 18 to 36 with a mean age of24 Fourteen men and twelve women completed theexperiment Results from four other subjects had to beexcluded from our data because the subjects becamelost in one or more of the cities All subjects were com-puter literate but with varying degrees of experiencewith VEs

69 Neuropsychological Tests

To provide an independent measure of a subset ofthe subjectsrsquo cognitive abilities subjects were tested in aseparate session using two standardized subtests fromthe Wechsler Adult Intelligent Scale-Revised (WAIS-R)(Wechsler 1955) The Vocabulary subtest comprisingpart of the WAIS-R Verbal subtests was used to assessthe subjectrsquos verbal ability Performance was scored ac-cording to standardized procedures based on the sub-jectrsquos degree of familiarity with a list of up to 35 wordsand their elegance of expression The second test wasthe Block Design subtest comprising part of theWAIS-R Performance subtests which was used to assessthe subjectsrsquo constructional-spatial ability and was ad-ministered in the standardized manner

610 Analysis

A one-way repeated measures analysis of variance(ANOVA) was performed for each of the dependentvariables and the overall times and distances The with-in-subjects variable was the landmark description typewith three levels text image and worldlet Post hocanalyses were completed using the Scheffe test for dif-ferences in sample means Correlation analysis was per-

formed across the scores of the neuropsychological testsand the scores of the waynding tests We adopted asignicance level of 005 on all analyses unless otherwisenoted

7 Results

Our primary result was that people spend moretime consulting the worldlet guidebook than the otherguidebooks but that they require less time to reach thegoal when using the worldlet guidebook The followingspecic ndings tell us more about the component ac-tivities that benetted from worldlet use

71 Time and Distance Measures

Figures 8 through 11 show time and distancemeasures in which the independent variable (type ofguidebook) is displayed as white gray and black barsalong with the standard error of the mean (SEM) Fig-ure 8 shows the mean overall time and distance requiredto reach the goal kiosk and gure 9 breaks the overalltime into its components Figure 10 gives the forwardand backtracking distance components Figure 11 showsthe mean number of times that each type of guidebook

Figure 8 (left) Mean total time The overall time spent traveling to

the goal was signicantly less using the worldlet guidebook than when

using the text (F(1 24) 5 989 p 001) or image (F(1 24) 5

463 p 001) guidebooks (right) Mean total distance The overall

distance traveled using the worldlet guidebook was signicantly less

than the distance traveled using the text (F(1 24) 5 1141 p

001) or image (F(1 24) 5 893 p 001) guidebooks

576 PRESENCE VOLUME 10 NUMBER 6

was opened Figure 12 diagrams the subjectsrsquo scores onthe Wechsler neuropsychological tests

Analysis of variance followed by Scheffe post hoc testswere performed for the different measures Table 2summarizes these results

72 Street Plots

Figure 13 shows plots of the streets traversed by the26 subjects in the three test cities The thickness andbrightness of each street line indicate the total number oftimes that the street was traversed For example widerbrighter streets indicate heavily used avenues Backtrackingcan increase the number of street traversals beyond theexpected minimum of 26 The starting point and goalkiosk are indicated on each plot and landmarks enroute tothe goal are marked with a star (q )

Figure 10 The mean distance moving forward while using the

worldlet guidebook is signicantly less than the forward distance

covered while using the text or image guidebook Backtracking

distance is almost nonexistent for the worldlet guidebook task

Figure 9 Turning looking for landmark is time spent standing in

one location Turning changing forward direction is time spent

changing forward direction Consulting guidebook is time spent with

the guidebook on screen Moving forward is time spent traveling over

new ground and Backtracking is time spent retreating Note that the

time spent consulting the worldlet guidebook is signicantly greater

than the time spent consulting the text or image guidebook Also note

that the time spent backtracking is near zero when using the worldlet

guidebook

Figure 11 Subjects opened the text and image guidebooks an

average of fteen to sixteen times while seeking the eight landmarks

enroute to the goal kiosk The worldlet guidebook was opened eleven

times on average

Figure 12 Z-transformed spatial and verbal scores from the

neuropsychological tests The mean raw scores were 45 and 55 for

the spatial and verbal tests respectively

Elvins et al 577

73 Subjective Measures

As part of the written questionnaire administeredafter the experiment subjects were asked how difcultthe search task was using each of the different types ofguidebooks Although responses were scattered a cleartrend in favor of the worldlet guidebook was evident asshown in table 3

74 Correlation Analysis

One surprising nding was a high inverse corre-lation ( 2 077 p 5 001) between spatial skills andthe time spent consulting the text guidebook Sub-jects with better spatial ability seemed able to form amental picture of a landmark based on textual de-scription more quickly than did subjects with poorer

spatial ability This is not the case for image andworldlet guidebooks however

We expected to nd a negative correlation betweenspatial skills and overall travel time in the image andworldlet guidebook cities The data did not supportthis nor did they support another of our hypothesesthat there would be a positive correlation between ver-bal skills and the performance in the text guidebookcity Verbal skills made no difference in overall waynd-ing time

75 Observations on Subject Strategies

Many subjects said that the tasks were harderthan they had imagined and afterwards several saidthat they felt tired No subjects complained of mo-

Table 2 Our ANOVA Indicated Signicant Differences between All the Sample Means and Post Hoc Analysis Was Warranted

ANOVA

Post hoc comparisons (Scheffe F-test)

text vs image text vs worldlet image vs worldlet

TimeConsulting guidebook 5559 p 00001 1966 p 0001 884 p 001 5488 p 0011Looking for landmark 2161 p 00001 047 1869 p 0001 1325 p 001Changing heading 1736 p 00001 002 1353 p 001 1249 p 001Moving forward 1602 p 00001 0003 1220 p 001 1183 p 001Backtracking 955 p 00003 022 826 p 001 586 p 001Overall 1034 p 00002 098 989 p 001 463 p 001

DistanceMoving forward 1533 p 00001 005 1223 p 001 1072 p 001Backtracking 1019 p 00002 023 884 p 001 621 p 001Overall 1365 p 00001 015 1141 p 001 893 p 001

GuidebookOpened 1527 p 00001 104 759 p 001 1426 p 0001

Post hoc (Scheffe) analysis indicated for all dependent variables signicant differences in the sample means of the textversus worldlet performance and image versus worldlet performance Performance using the text versus imageguidebooks was not signicantly different except that the time spent consulting the text guidebook was signicantlygreater than the time spent consulting the image guidebook In each table cell the rst number is the F value and thesecond number is the p value Asterisk indicates signicance p 005 Time and distance measures are explained in thecaptions for gures 8 through 11

578 PRESENCE VOLUME 10 NUMBER 6

tion sickness Most subjects required 70 min or moreto nish all aspects of the computer-based part of theexperiment

We observed subjects to use three general types ofstrategies while deciding what direction to go from astreet intersection

c Look down each street from many viewpointswithin the intersection until a landmark or land-mark context is seen

c Look down each street and if nothing looks famil-iar search exhaustively in each direction until thenext landmark is found

c Look down each street and if nothing looks famil-iar try going one-half block in each direction andlook again

Subjects using the rst strategy seemed to nd theirway more quickly and with less backtracking than didsubjects who used the second and third strategies An-other strategy described by some subjects on their writ-ten questionnaire involved viewing a landmark in theworldlet guidebook from many distant viewpoints in thehopes that one of the viewpoints would be similar to thedirection from which they would actually approach the

Figure 13 In these street plots line thickness and brightness

indicates the number of times that a street was traversed In top and

middle the 26 subjects traversed some streets 47 times indicating

backtracking Top and middle also exhibit circuitous trails In bottom

very little backtracking occurred and circuitous trails are minimal

Table 3 Subjects Were Asked to Rate the Difculty of theSearch Task Using the Three Different Types of Guidebooks

Text Image Worldlet

Very easy 0 2 11Easy 10 8 9Doable 10 8 6Difcult 5 7 0Very difcult 1 1 0

Median Doable Doable Easy

Elvins et al 579

landmark in the city We observed that some subjectsoften forgot what it was they were looking for and sohad to open the guidebook momentarily to remindthemselves

Not surprisingly landmarks with prominent signage(such as ldquoBobrsquos Groceryrdquo) and landmarks painted withmore highly saturated colors seemed easier to nd thanother types of landmarks Also after passing all thelandmarks several subjects exclaimed that they had lostall sense of direction within the world

A nal unexpected result were the many similarities be-tween the text and image guidebooks The text and imageguidebooks were only signicantly different in the timesubjects spent consulting them Considering that text andimages represent landmarks in very different ways we ex-pected differences in time and distance measures

76 Major Findings

In interpreting these ndings it is important toremember that the subjects performed the wayndingtasks using only the landmark knowledge gained bystudying the guidebooks No procedural or survey(map) information was provided

c As we predicted the total time spent traveling tothe goal was signicantly less using the worldletguidebook than when using the text or imageguidebooks

c Subjects spent signicantly more time studying theworldlet guidebook than they did the text or imageguidebook

c Increased worldlet study time was more than com-pensated for by a reduction in the time required toreach the goal Using the worldlet guidebook sub-jects traveled an average of 55 faster than whenusing the text guidebook and 38 faster than whenusing the image guidebook

c Increased worldlet study time also allowed subjectsto move more directly to the goal and reducedbacktracking time to almost zero This is showndiagramatically in gure 13 in which gures 13aand 13b show brighter wider streets than in gure

13c Figure 13c also exhibits fewer detours and lesswandering than do gures 13a and 13b

We postulate that compared to the landmark knowl-edge gained from text and image guidebooks the im-proved landmark knowledge built while consulting theworldlet guidebook enabled subjects to refer less fre-quently to the guidebook This postulation is furthersupported by examining worldlet guidebook results

c Subjects spent signicantly more time studying theworldlet guidebook than they did the other types ofguidebooks

c When using the worldlet guidebook subjects trav-eled faster more directly and with more skill tothe goal kiosk

c Subjectsrsquo median rating for the difculty of thesearch task was Easy when using the worldlet guide-book

8 Conclusion

The waynding literature shows that landmarksare critical in navigating an environment whether it isreal or virtual While building landmark knowledgetravelers characterize a landmark representation by its3-D shape size texture context and so forth In awaynding task textual or image guidebook landmarkdescriptions fail to engage the full range of 3-D land-mark characteristics recognized and used by travelers tond their way

We know of no studies that have considered the com-parative advantages of different types of guidebooks forwaynding Intuitively one would predict that 3-Dthumbnails make better guidebooks for 3-D worldsthan do text and 2-D thumbnail images and indeedthis is what we found We wanted however to identifythe component activities in using guidebooks that mostbenetted from 3-D elements Perhaps of greatest inter-est was our nding that worldlets enabled subjects tond their way in complex environments with almost nobacktracking By spending more time studying theworldlets in their guidebooks they spent less time look-ing around reorienting themselves conducting exhaus-

580 PRESENCE VOLUME 10 NUMBER 6

tive searches and performing similar time-wasting activ-ities We must be cautious however in what weconclude because worldlets contain more informationthan do text or 2-D thumbnail images To control forthis difference in information content more experi-ments are necessary

9 Future Work

We would like to investigate the comparative ad-vantages of worldlet guidebooks for waynding tasks incorresponding virtual and real world environments(both urban and rural) We are also considering an ex-periment to evaluate tandem use of traditional 2-Dmaps and landmark guidebooks Requiring effectivemap and compass use in a natural environment thesport of orienteering (Kjellstrom 1976) might benetfrom worldlets By encapsulating the region around theagged control points into a worldlet orienteers couldgain rst-person experience with natural landmarks suchas foliage rocks gullies shadows and so on beforebeginning the search

Acknowledgments

Our appreciation goes to Planet 9 Studios (wwwplanet9com)for the use of their city building textures Thanks to JohnMoreland and Mike Bailey for technical assistance on thisproject and thanks to Suzi Feeney for collaborating in thepilot study design UCSD librarian Debbie Cox and her staffwere invaluable in locating references Thanks also to JohnHelly and Reagan Moore for their support San Diego Super-computer Center is funded by NSF the University of Califor-nia the State of California and industrial partners

References

Appleyard D A (1976) Planning a Pluralistic City Con-icting Realities in Ciudad Guayana Cambridge MAMIT Press

Art1 Com (1997) The cybercity Berlin project [on-line]Available httpwwwartcomdeprojects

Chen S E (1995) QuickTime VRmdashAn image-based ap-proach to VE navigation Proceedings of SIGGRAPH 9529 ndash38

Darken R P amp Sibert J L (1996) Waynding strategiesand behaviors in large virtual worlds Proceedings of CHI 96142ndash149

Downs R J amp Stea D (1973) Cognitive maps and spatialbehavior In R J Downs amp D Stea (Eds) Image and En-vironment Chicago Aldine Publishing Company 8 ndash26

Elvins T T (1997) Waynding in layered informationworlds Proceedings of the CODATA Euro-American Work-shop on Data and Information Visualization 6 ndash7

Elvins T T Nadeau D R amp Kirsh D (1997) World-letsmdash3D thumbnails for waynding in virtual environmentsProceedings of the ACM User Interface Software and Technol-ogy Symposium 21ndash30

Elvins T T (1998) WorldletsmdashA novel 3D thumbnail af-fordance for waynding in large virtual environmentsUnpublished doctoral dissertation Department of Elec-trical and Computer Engineering University of Califor-niandashSan Diego

Elvins T T Nadeau D R Kirsh D amp Schul R (1998)Worldletsmdash3D thumbnails for 3D browsing Proceedings ofthe ACM CHI Conference 21ndash30

Foley J van Dam A Feiner S amp Hughes J (1990) Com-puter Graphics Principles and Practice Reading MA Addi-son-Wesley

Gillner S amp Mallot H A (1998) Navigation and acquisi-tion of spatial knowledge in a virtual maze Journal of Cog-nitive Neuroscience 10(4) 445ndash 463

Goldin S E amp Thorndyke P W (1982) Simulating naviga-tion for spatial knowledge acquisition Human Factors24(4) 457ndash 471

ISO (1997) The Virtual Reality Modeling Language Inter-national Standard ISOIEC 14772-11997 [on-line] Avail-able httpwwwvrmlorgSpecicationsVRML97

Kjellstrom B (1976) Be Expert with Map and Compass TheComplete Orienteering Handbook New York Scribner Pub-lishers

Liggett R Friedman S amp Jepson W (1995) Interactivedesigndecision making in a virtual urban world visual sim-ulation and GIS Proceedings of the 1995 ESRI User Confer-ence Environmental Systems Research [on-line] Availablehttpwwwesricombasecommonuserconfproc95

Lynch K (1960) The Image of the City Cambridge MAMIT Press

Elvins et al 581

was opened Figure 12 diagrams the subjectsrsquo scores onthe Wechsler neuropsychological tests

Analysis of variance followed by Scheffe post hoc testswere performed for the different measures Table 2summarizes these results

72 Street Plots

Figure 13 shows plots of the streets traversed by the26 subjects in the three test cities The thickness andbrightness of each street line indicate the total number oftimes that the street was traversed For example widerbrighter streets indicate heavily used avenues Backtrackingcan increase the number of street traversals beyond theexpected minimum of 26 The starting point and goalkiosk are indicated on each plot and landmarks enroute tothe goal are marked with a star (q )

Figure 10 The mean distance moving forward while using the

worldlet guidebook is signicantly less than the forward distance

covered while using the text or image guidebook Backtracking

distance is almost nonexistent for the worldlet guidebook task

Figure 9 Turning looking for landmark is time spent standing in

one location Turning changing forward direction is time spent

changing forward direction Consulting guidebook is time spent with

the guidebook on screen Moving forward is time spent traveling over

new ground and Backtracking is time spent retreating Note that the

time spent consulting the worldlet guidebook is signicantly greater

than the time spent consulting the text or image guidebook Also note

that the time spent backtracking is near zero when using the worldlet

guidebook

Figure 11 Subjects opened the text and image guidebooks an

average of fteen to sixteen times while seeking the eight landmarks

enroute to the goal kiosk The worldlet guidebook was opened eleven

times on average

Figure 12 Z-transformed spatial and verbal scores from the

neuropsychological tests The mean raw scores were 45 and 55 for

the spatial and verbal tests respectively

Elvins et al 577

73 Subjective Measures

As part of the written questionnaire administeredafter the experiment subjects were asked how difcultthe search task was using each of the different types ofguidebooks Although responses were scattered a cleartrend in favor of the worldlet guidebook was evident asshown in table 3

74 Correlation Analysis

One surprising nding was a high inverse corre-lation ( 2 077 p 5 001) between spatial skills andthe time spent consulting the text guidebook Sub-jects with better spatial ability seemed able to form amental picture of a landmark based on textual de-scription more quickly than did subjects with poorer

spatial ability This is not the case for image andworldlet guidebooks however

We expected to nd a negative correlation betweenspatial skills and overall travel time in the image andworldlet guidebook cities The data did not supportthis nor did they support another of our hypothesesthat there would be a positive correlation between ver-bal skills and the performance in the text guidebookcity Verbal skills made no difference in overall waynd-ing time

75 Observations on Subject Strategies

Many subjects said that the tasks were harderthan they had imagined and afterwards several saidthat they felt tired No subjects complained of mo-

Table 2 Our ANOVA Indicated Signicant Differences between All the Sample Means and Post Hoc Analysis Was Warranted

ANOVA

Post hoc comparisons (Scheffe F-test)

text vs image text vs worldlet image vs worldlet

TimeConsulting guidebook 5559 p 00001 1966 p 0001 884 p 001 5488 p 0011Looking for landmark 2161 p 00001 047 1869 p 0001 1325 p 001Changing heading 1736 p 00001 002 1353 p 001 1249 p 001Moving forward 1602 p 00001 0003 1220 p 001 1183 p 001Backtracking 955 p 00003 022 826 p 001 586 p 001Overall 1034 p 00002 098 989 p 001 463 p 001

DistanceMoving forward 1533 p 00001 005 1223 p 001 1072 p 001Backtracking 1019 p 00002 023 884 p 001 621 p 001Overall 1365 p 00001 015 1141 p 001 893 p 001

GuidebookOpened 1527 p 00001 104 759 p 001 1426 p 0001

Post hoc (Scheffe) analysis indicated for all dependent variables signicant differences in the sample means of the textversus worldlet performance and image versus worldlet performance Performance using the text versus imageguidebooks was not signicantly different except that the time spent consulting the text guidebook was signicantlygreater than the time spent consulting the image guidebook In each table cell the rst number is the F value and thesecond number is the p value Asterisk indicates signicance p 005 Time and distance measures are explained in thecaptions for gures 8 through 11

578 PRESENCE VOLUME 10 NUMBER 6

tion sickness Most subjects required 70 min or moreto nish all aspects of the computer-based part of theexperiment

We observed subjects to use three general types ofstrategies while deciding what direction to go from astreet intersection

c Look down each street from many viewpointswithin the intersection until a landmark or land-mark context is seen

c Look down each street and if nothing looks famil-iar search exhaustively in each direction until thenext landmark is found

c Look down each street and if nothing looks famil-iar try going one-half block in each direction andlook again

Subjects using the rst strategy seemed to nd theirway more quickly and with less backtracking than didsubjects who used the second and third strategies An-other strategy described by some subjects on their writ-ten questionnaire involved viewing a landmark in theworldlet guidebook from many distant viewpoints in thehopes that one of the viewpoints would be similar to thedirection from which they would actually approach the

Figure 13 In these street plots line thickness and brightness

indicates the number of times that a street was traversed In top and

middle the 26 subjects traversed some streets 47 times indicating

backtracking Top and middle also exhibit circuitous trails In bottom

very little backtracking occurred and circuitous trails are minimal

Table 3 Subjects Were Asked to Rate the Difculty of theSearch Task Using the Three Different Types of Guidebooks

Text Image Worldlet

Very easy 0 2 11Easy 10 8 9Doable 10 8 6Difcult 5 7 0Very difcult 1 1 0

Median Doable Doable Easy

Elvins et al 579

landmark in the city We observed that some subjectsoften forgot what it was they were looking for and sohad to open the guidebook momentarily to remindthemselves

Not surprisingly landmarks with prominent signage(such as ldquoBobrsquos Groceryrdquo) and landmarks painted withmore highly saturated colors seemed easier to nd thanother types of landmarks Also after passing all thelandmarks several subjects exclaimed that they had lostall sense of direction within the world

A nal unexpected result were the many similarities be-tween the text and image guidebooks The text and imageguidebooks were only signicantly different in the timesubjects spent consulting them Considering that text andimages represent landmarks in very different ways we ex-pected differences in time and distance measures

76 Major Findings

In interpreting these ndings it is important toremember that the subjects performed the wayndingtasks using only the landmark knowledge gained bystudying the guidebooks No procedural or survey(map) information was provided

c As we predicted the total time spent traveling tothe goal was signicantly less using the worldletguidebook than when using the text or imageguidebooks

c Subjects spent signicantly more time studying theworldlet guidebook than they did the text or imageguidebook

c Increased worldlet study time was more than com-pensated for by a reduction in the time required toreach the goal Using the worldlet guidebook sub-jects traveled an average of 55 faster than whenusing the text guidebook and 38 faster than whenusing the image guidebook

c Increased worldlet study time also allowed subjectsto move more directly to the goal and reducedbacktracking time to almost zero This is showndiagramatically in gure 13 in which gures 13aand 13b show brighter wider streets than in gure

13c Figure 13c also exhibits fewer detours and lesswandering than do gures 13a and 13b

We postulate that compared to the landmark knowl-edge gained from text and image guidebooks the im-proved landmark knowledge built while consulting theworldlet guidebook enabled subjects to refer less fre-quently to the guidebook This postulation is furthersupported by examining worldlet guidebook results

c Subjects spent signicantly more time studying theworldlet guidebook than they did the other types ofguidebooks

c When using the worldlet guidebook subjects trav-eled faster more directly and with more skill tothe goal kiosk

c Subjectsrsquo median rating for the difculty of thesearch task was Easy when using the worldlet guide-book

8 Conclusion

The waynding literature shows that landmarksare critical in navigating an environment whether it isreal or virtual While building landmark knowledgetravelers characterize a landmark representation by its3-D shape size texture context and so forth In awaynding task textual or image guidebook landmarkdescriptions fail to engage the full range of 3-D land-mark characteristics recognized and used by travelers tond their way

We know of no studies that have considered the com-parative advantages of different types of guidebooks forwaynding Intuitively one would predict that 3-Dthumbnails make better guidebooks for 3-D worldsthan do text and 2-D thumbnail images and indeedthis is what we found We wanted however to identifythe component activities in using guidebooks that mostbenetted from 3-D elements Perhaps of greatest inter-est was our nding that worldlets enabled subjects tond their way in complex environments with almost nobacktracking By spending more time studying theworldlets in their guidebooks they spent less time look-ing around reorienting themselves conducting exhaus-

580 PRESENCE VOLUME 10 NUMBER 6

tive searches and performing similar time-wasting activ-ities We must be cautious however in what weconclude because worldlets contain more informationthan do text or 2-D thumbnail images To control forthis difference in information content more experi-ments are necessary

9 Future Work

We would like to investigate the comparative ad-vantages of worldlet guidebooks for waynding tasks incorresponding virtual and real world environments(both urban and rural) We are also considering an ex-periment to evaluate tandem use of traditional 2-Dmaps and landmark guidebooks Requiring effectivemap and compass use in a natural environment thesport of orienteering (Kjellstrom 1976) might benetfrom worldlets By encapsulating the region around theagged control points into a worldlet orienteers couldgain rst-person experience with natural landmarks suchas foliage rocks gullies shadows and so on beforebeginning the search

Acknowledgments

Our appreciation goes to Planet 9 Studios (wwwplanet9com)for the use of their city building textures Thanks to JohnMoreland and Mike Bailey for technical assistance on thisproject and thanks to Suzi Feeney for collaborating in thepilot study design UCSD librarian Debbie Cox and her staffwere invaluable in locating references Thanks also to JohnHelly and Reagan Moore for their support San Diego Super-computer Center is funded by NSF the University of Califor-nia the State of California and industrial partners

References

Appleyard D A (1976) Planning a Pluralistic City Con-icting Realities in Ciudad Guayana Cambridge MAMIT Press

Art1 Com (1997) The cybercity Berlin project [on-line]Available httpwwwartcomdeprojects

Chen S E (1995) QuickTime VRmdashAn image-based ap-proach to VE navigation Proceedings of SIGGRAPH 9529 ndash38

Darken R P amp Sibert J L (1996) Waynding strategiesand behaviors in large virtual worlds Proceedings of CHI 96142ndash149

Downs R J amp Stea D (1973) Cognitive maps and spatialbehavior In R J Downs amp D Stea (Eds) Image and En-vironment Chicago Aldine Publishing Company 8 ndash26

Elvins T T (1997) Waynding in layered informationworlds Proceedings of the CODATA Euro-American Work-shop on Data and Information Visualization 6 ndash7

Elvins T T Nadeau D R amp Kirsh D (1997) World-letsmdash3D thumbnails for waynding in virtual environmentsProceedings of the ACM User Interface Software and Technol-ogy Symposium 21ndash30

Elvins T T (1998) WorldletsmdashA novel 3D thumbnail af-fordance for waynding in large virtual environmentsUnpublished doctoral dissertation Department of Elec-trical and Computer Engineering University of Califor-niandashSan Diego

Elvins T T Nadeau D R Kirsh D amp Schul R (1998)Worldletsmdash3D thumbnails for 3D browsing Proceedings ofthe ACM CHI Conference 21ndash30

Foley J van Dam A Feiner S amp Hughes J (1990) Com-puter Graphics Principles and Practice Reading MA Addi-son-Wesley

Gillner S amp Mallot H A (1998) Navigation and acquisi-tion of spatial knowledge in a virtual maze Journal of Cog-nitive Neuroscience 10(4) 445ndash 463

Goldin S E amp Thorndyke P W (1982) Simulating naviga-tion for spatial knowledge acquisition Human Factors24(4) 457ndash 471

ISO (1997) The Virtual Reality Modeling Language Inter-national Standard ISOIEC 14772-11997 [on-line] Avail-able httpwwwvrmlorgSpecicationsVRML97

Kjellstrom B (1976) Be Expert with Map and Compass TheComplete Orienteering Handbook New York Scribner Pub-lishers

Liggett R Friedman S amp Jepson W (1995) Interactivedesigndecision making in a virtual urban world visual sim-ulation and GIS Proceedings of the 1995 ESRI User Confer-ence Environmental Systems Research [on-line] Availablehttpwwwesricombasecommonuserconfproc95

Lynch K (1960) The Image of the City Cambridge MAMIT Press

Elvins et al 581

73 Subjective Measures

As part of the written questionnaire administeredafter the experiment subjects were asked how difcultthe search task was using each of the different types ofguidebooks Although responses were scattered a cleartrend in favor of the worldlet guidebook was evident asshown in table 3

74 Correlation Analysis

One surprising nding was a high inverse corre-lation ( 2 077 p 5 001) between spatial skills andthe time spent consulting the text guidebook Sub-jects with better spatial ability seemed able to form amental picture of a landmark based on textual de-scription more quickly than did subjects with poorer

spatial ability This is not the case for image andworldlet guidebooks however

We expected to nd a negative correlation betweenspatial skills and overall travel time in the image andworldlet guidebook cities The data did not supportthis nor did they support another of our hypothesesthat there would be a positive correlation between ver-bal skills and the performance in the text guidebookcity Verbal skills made no difference in overall waynd-ing time

75 Observations on Subject Strategies

Many subjects said that the tasks were harderthan they had imagined and afterwards several saidthat they felt tired No subjects complained of mo-

Table 2 Our ANOVA Indicated Signicant Differences between All the Sample Means and Post Hoc Analysis Was Warranted

ANOVA

Post hoc comparisons (Scheffe F-test)

text vs image text vs worldlet image vs worldlet

TimeConsulting guidebook 5559 p 00001 1966 p 0001 884 p 001 5488 p 0011Looking for landmark 2161 p 00001 047 1869 p 0001 1325 p 001Changing heading 1736 p 00001 002 1353 p 001 1249 p 001Moving forward 1602 p 00001 0003 1220 p 001 1183 p 001Backtracking 955 p 00003 022 826 p 001 586 p 001Overall 1034 p 00002 098 989 p 001 463 p 001

DistanceMoving forward 1533 p 00001 005 1223 p 001 1072 p 001Backtracking 1019 p 00002 023 884 p 001 621 p 001Overall 1365 p 00001 015 1141 p 001 893 p 001

GuidebookOpened 1527 p 00001 104 759 p 001 1426 p 0001

Post hoc (Scheffe) analysis indicated for all dependent variables signicant differences in the sample means of the textversus worldlet performance and image versus worldlet performance Performance using the text versus imageguidebooks was not signicantly different except that the time spent consulting the text guidebook was signicantlygreater than the time spent consulting the image guidebook In each table cell the rst number is the F value and thesecond number is the p value Asterisk indicates signicance p 005 Time and distance measures are explained in thecaptions for gures 8 through 11

578 PRESENCE VOLUME 10 NUMBER 6

tion sickness Most subjects required 70 min or moreto nish all aspects of the computer-based part of theexperiment

We observed subjects to use three general types ofstrategies while deciding what direction to go from astreet intersection

c Look down each street from many viewpointswithin the intersection until a landmark or land-mark context is seen

c Look down each street and if nothing looks famil-iar search exhaustively in each direction until thenext landmark is found

c Look down each street and if nothing looks famil-iar try going one-half block in each direction andlook again

Subjects using the rst strategy seemed to nd theirway more quickly and with less backtracking than didsubjects who used the second and third strategies An-other strategy described by some subjects on their writ-ten questionnaire involved viewing a landmark in theworldlet guidebook from many distant viewpoints in thehopes that one of the viewpoints would be similar to thedirection from which they would actually approach the

Figure 13 In these street plots line thickness and brightness

indicates the number of times that a street was traversed In top and

middle the 26 subjects traversed some streets 47 times indicating

backtracking Top and middle also exhibit circuitous trails In bottom

very little backtracking occurred and circuitous trails are minimal

Table 3 Subjects Were Asked to Rate the Difculty of theSearch Task Using the Three Different Types of Guidebooks

Text Image Worldlet

Very easy 0 2 11Easy 10 8 9Doable 10 8 6Difcult 5 7 0Very difcult 1 1 0

Median Doable Doable Easy

Elvins et al 579

landmark in the city We observed that some subjectsoften forgot what it was they were looking for and sohad to open the guidebook momentarily to remindthemselves

Not surprisingly landmarks with prominent signage(such as ldquoBobrsquos Groceryrdquo) and landmarks painted withmore highly saturated colors seemed easier to nd thanother types of landmarks Also after passing all thelandmarks several subjects exclaimed that they had lostall sense of direction within the world

A nal unexpected result were the many similarities be-tween the text and image guidebooks The text and imageguidebooks were only signicantly different in the timesubjects spent consulting them Considering that text andimages represent landmarks in very different ways we ex-pected differences in time and distance measures

76 Major Findings

In interpreting these ndings it is important toremember that the subjects performed the wayndingtasks using only the landmark knowledge gained bystudying the guidebooks No procedural or survey(map) information was provided

c As we predicted the total time spent traveling tothe goal was signicantly less using the worldletguidebook than when using the text or imageguidebooks

c Subjects spent signicantly more time studying theworldlet guidebook than they did the text or imageguidebook

c Increased worldlet study time was more than com-pensated for by a reduction in the time required toreach the goal Using the worldlet guidebook sub-jects traveled an average of 55 faster than whenusing the text guidebook and 38 faster than whenusing the image guidebook

c Increased worldlet study time also allowed subjectsto move more directly to the goal and reducedbacktracking time to almost zero This is showndiagramatically in gure 13 in which gures 13aand 13b show brighter wider streets than in gure

13c Figure 13c also exhibits fewer detours and lesswandering than do gures 13a and 13b

We postulate that compared to the landmark knowl-edge gained from text and image guidebooks the im-proved landmark knowledge built while consulting theworldlet guidebook enabled subjects to refer less fre-quently to the guidebook This postulation is furthersupported by examining worldlet guidebook results

c Subjects spent signicantly more time studying theworldlet guidebook than they did the other types ofguidebooks

c When using the worldlet guidebook subjects trav-eled faster more directly and with more skill tothe goal kiosk

c Subjectsrsquo median rating for the difculty of thesearch task was Easy when using the worldlet guide-book

8 Conclusion

The waynding literature shows that landmarksare critical in navigating an environment whether it isreal or virtual While building landmark knowledgetravelers characterize a landmark representation by its3-D shape size texture context and so forth In awaynding task textual or image guidebook landmarkdescriptions fail to engage the full range of 3-D land-mark characteristics recognized and used by travelers tond their way

We know of no studies that have considered the com-parative advantages of different types of guidebooks forwaynding Intuitively one would predict that 3-Dthumbnails make better guidebooks for 3-D worldsthan do text and 2-D thumbnail images and indeedthis is what we found We wanted however to identifythe component activities in using guidebooks that mostbenetted from 3-D elements Perhaps of greatest inter-est was our nding that worldlets enabled subjects tond their way in complex environments with almost nobacktracking By spending more time studying theworldlets in their guidebooks they spent less time look-ing around reorienting themselves conducting exhaus-

580 PRESENCE VOLUME 10 NUMBER 6

tive searches and performing similar time-wasting activ-ities We must be cautious however in what weconclude because worldlets contain more informationthan do text or 2-D thumbnail images To control forthis difference in information content more experi-ments are necessary

9 Future Work

We would like to investigate the comparative ad-vantages of worldlet guidebooks for waynding tasks incorresponding virtual and real world environments(both urban and rural) We are also considering an ex-periment to evaluate tandem use of traditional 2-Dmaps and landmark guidebooks Requiring effectivemap and compass use in a natural environment thesport of orienteering (Kjellstrom 1976) might benetfrom worldlets By encapsulating the region around theagged control points into a worldlet orienteers couldgain rst-person experience with natural landmarks suchas foliage rocks gullies shadows and so on beforebeginning the search

Acknowledgments

Our appreciation goes to Planet 9 Studios (wwwplanet9com)for the use of their city building textures Thanks to JohnMoreland and Mike Bailey for technical assistance on thisproject and thanks to Suzi Feeney for collaborating in thepilot study design UCSD librarian Debbie Cox and her staffwere invaluable in locating references Thanks also to JohnHelly and Reagan Moore for their support San Diego Super-computer Center is funded by NSF the University of Califor-nia the State of California and industrial partners

References

Appleyard D A (1976) Planning a Pluralistic City Con-icting Realities in Ciudad Guayana Cambridge MAMIT Press

Art1 Com (1997) The cybercity Berlin project [on-line]Available httpwwwartcomdeprojects

Chen S E (1995) QuickTime VRmdashAn image-based ap-proach to VE navigation Proceedings of SIGGRAPH 9529 ndash38

Darken R P amp Sibert J L (1996) Waynding strategiesand behaviors in large virtual worlds Proceedings of CHI 96142ndash149

Downs R J amp Stea D (1973) Cognitive maps and spatialbehavior In R J Downs amp D Stea (Eds) Image and En-vironment Chicago Aldine Publishing Company 8 ndash26

Elvins T T (1997) Waynding in layered informationworlds Proceedings of the CODATA Euro-American Work-shop on Data and Information Visualization 6 ndash7

Elvins T T Nadeau D R amp Kirsh D (1997) World-letsmdash3D thumbnails for waynding in virtual environmentsProceedings of the ACM User Interface Software and Technol-ogy Symposium 21ndash30

Elvins T T (1998) WorldletsmdashA novel 3D thumbnail af-fordance for waynding in large virtual environmentsUnpublished doctoral dissertation Department of Elec-trical and Computer Engineering University of Califor-niandashSan Diego

Elvins T T Nadeau D R Kirsh D amp Schul R (1998)Worldletsmdash3D thumbnails for 3D browsing Proceedings ofthe ACM CHI Conference 21ndash30

Foley J van Dam A Feiner S amp Hughes J (1990) Com-puter Graphics Principles and Practice Reading MA Addi-son-Wesley

Gillner S amp Mallot H A (1998) Navigation and acquisi-tion of spatial knowledge in a virtual maze Journal of Cog-nitive Neuroscience 10(4) 445ndash 463

Goldin S E amp Thorndyke P W (1982) Simulating naviga-tion for spatial knowledge acquisition Human Factors24(4) 457ndash 471

ISO (1997) The Virtual Reality Modeling Language Inter-national Standard ISOIEC 14772-11997 [on-line] Avail-able httpwwwvrmlorgSpecicationsVRML97

Kjellstrom B (1976) Be Expert with Map and Compass TheComplete Orienteering Handbook New York Scribner Pub-lishers

Liggett R Friedman S amp Jepson W (1995) Interactivedesigndecision making in a virtual urban world visual sim-ulation and GIS Proceedings of the 1995 ESRI User Confer-ence Environmental Systems Research [on-line] Availablehttpwwwesricombasecommonuserconfproc95

Lynch K (1960) The Image of the City Cambridge MAMIT Press

Elvins et al 581

tion sickness Most subjects required 70 min or moreto nish all aspects of the computer-based part of theexperiment

We observed subjects to use three general types ofstrategies while deciding what direction to go from astreet intersection

c Look down each street from many viewpointswithin the intersection until a landmark or land-mark context is seen

c Look down each street and if nothing looks famil-iar search exhaustively in each direction until thenext landmark is found

c Look down each street and if nothing looks famil-iar try going one-half block in each direction andlook again

Subjects using the rst strategy seemed to nd theirway more quickly and with less backtracking than didsubjects who used the second and third strategies An-other strategy described by some subjects on their writ-ten questionnaire involved viewing a landmark in theworldlet guidebook from many distant viewpoints in thehopes that one of the viewpoints would be similar to thedirection from which they would actually approach the

Figure 13 In these street plots line thickness and brightness

indicates the number of times that a street was traversed In top and

middle the 26 subjects traversed some streets 47 times indicating

backtracking Top and middle also exhibit circuitous trails In bottom

very little backtracking occurred and circuitous trails are minimal

Table 3 Subjects Were Asked to Rate the Difculty of theSearch Task Using the Three Different Types of Guidebooks

Text Image Worldlet

Very easy 0 2 11Easy 10 8 9Doable 10 8 6Difcult 5 7 0Very difcult 1 1 0

Median Doable Doable Easy

Elvins et al 579

landmark in the city We observed that some subjectsoften forgot what it was they were looking for and sohad to open the guidebook momentarily to remindthemselves

Not surprisingly landmarks with prominent signage(such as ldquoBobrsquos Groceryrdquo) and landmarks painted withmore highly saturated colors seemed easier to nd thanother types of landmarks Also after passing all thelandmarks several subjects exclaimed that they had lostall sense of direction within the world

A nal unexpected result were the many similarities be-tween the text and image guidebooks The text and imageguidebooks were only signicantly different in the timesubjects spent consulting them Considering that text andimages represent landmarks in very different ways we ex-pected differences in time and distance measures

76 Major Findings

In interpreting these ndings it is important toremember that the subjects performed the wayndingtasks using only the landmark knowledge gained bystudying the guidebooks No procedural or survey(map) information was provided

c As we predicted the total time spent traveling tothe goal was signicantly less using the worldletguidebook than when using the text or imageguidebooks

c Subjects spent signicantly more time studying theworldlet guidebook than they did the text or imageguidebook

c Increased worldlet study time was more than com-pensated for by a reduction in the time required toreach the goal Using the worldlet guidebook sub-jects traveled an average of 55 faster than whenusing the text guidebook and 38 faster than whenusing the image guidebook

c Increased worldlet study time also allowed subjectsto move more directly to the goal and reducedbacktracking time to almost zero This is showndiagramatically in gure 13 in which gures 13aand 13b show brighter wider streets than in gure

13c Figure 13c also exhibits fewer detours and lesswandering than do gures 13a and 13b

We postulate that compared to the landmark knowl-edge gained from text and image guidebooks the im-proved landmark knowledge built while consulting theworldlet guidebook enabled subjects to refer less fre-quently to the guidebook This postulation is furthersupported by examining worldlet guidebook results

c Subjects spent signicantly more time studying theworldlet guidebook than they did the other types ofguidebooks

c When using the worldlet guidebook subjects trav-eled faster more directly and with more skill tothe goal kiosk

c Subjectsrsquo median rating for the difculty of thesearch task was Easy when using the worldlet guide-book

8 Conclusion

The waynding literature shows that landmarksare critical in navigating an environment whether it isreal or virtual While building landmark knowledgetravelers characterize a landmark representation by its3-D shape size texture context and so forth In awaynding task textual or image guidebook landmarkdescriptions fail to engage the full range of 3-D land-mark characteristics recognized and used by travelers tond their way

We know of no studies that have considered the com-parative advantages of different types of guidebooks forwaynding Intuitively one would predict that 3-Dthumbnails make better guidebooks for 3-D worldsthan do text and 2-D thumbnail images and indeedthis is what we found We wanted however to identifythe component activities in using guidebooks that mostbenetted from 3-D elements Perhaps of greatest inter-est was our nding that worldlets enabled subjects tond their way in complex environments with almost nobacktracking By spending more time studying theworldlets in their guidebooks they spent less time look-ing around reorienting themselves conducting exhaus-

580 PRESENCE VOLUME 10 NUMBER 6

tive searches and performing similar time-wasting activ-ities We must be cautious however in what weconclude because worldlets contain more informationthan do text or 2-D thumbnail images To control forthis difference in information content more experi-ments are necessary

9 Future Work

We would like to investigate the comparative ad-vantages of worldlet guidebooks for waynding tasks incorresponding virtual and real world environments(both urban and rural) We are also considering an ex-periment to evaluate tandem use of traditional 2-Dmaps and landmark guidebooks Requiring effectivemap and compass use in a natural environment thesport of orienteering (Kjellstrom 1976) might benetfrom worldlets By encapsulating the region around theagged control points into a worldlet orienteers couldgain rst-person experience with natural landmarks suchas foliage rocks gullies shadows and so on beforebeginning the search

Acknowledgments

Our appreciation goes to Planet 9 Studios (wwwplanet9com)for the use of their city building textures Thanks to JohnMoreland and Mike Bailey for technical assistance on thisproject and thanks to Suzi Feeney for collaborating in thepilot study design UCSD librarian Debbie Cox and her staffwere invaluable in locating references Thanks also to JohnHelly and Reagan Moore for their support San Diego Super-computer Center is funded by NSF the University of Califor-nia the State of California and industrial partners

References

Appleyard D A (1976) Planning a Pluralistic City Con-icting Realities in Ciudad Guayana Cambridge MAMIT Press

Art1 Com (1997) The cybercity Berlin project [on-line]Available httpwwwartcomdeprojects

Chen S E (1995) QuickTime VRmdashAn image-based ap-proach to VE navigation Proceedings of SIGGRAPH 9529 ndash38

Darken R P amp Sibert J L (1996) Waynding strategiesand behaviors in large virtual worlds Proceedings of CHI 96142ndash149

Downs R J amp Stea D (1973) Cognitive maps and spatialbehavior In R J Downs amp D Stea (Eds) Image and En-vironment Chicago Aldine Publishing Company 8 ndash26

Elvins T T (1997) Waynding in layered informationworlds Proceedings of the CODATA Euro-American Work-shop on Data and Information Visualization 6 ndash7

Elvins T T Nadeau D R amp Kirsh D (1997) World-letsmdash3D thumbnails for waynding in virtual environmentsProceedings of the ACM User Interface Software and Technol-ogy Symposium 21ndash30

Elvins T T (1998) WorldletsmdashA novel 3D thumbnail af-fordance for waynding in large virtual environmentsUnpublished doctoral dissertation Department of Elec-trical and Computer Engineering University of Califor-niandashSan Diego

Elvins T T Nadeau D R Kirsh D amp Schul R (1998)Worldletsmdash3D thumbnails for 3D browsing Proceedings ofthe ACM CHI Conference 21ndash30

Foley J van Dam A Feiner S amp Hughes J (1990) Com-puter Graphics Principles and Practice Reading MA Addi-son-Wesley

Gillner S amp Mallot H A (1998) Navigation and acquisi-tion of spatial knowledge in a virtual maze Journal of Cog-nitive Neuroscience 10(4) 445ndash 463

Goldin S E amp Thorndyke P W (1982) Simulating naviga-tion for spatial knowledge acquisition Human Factors24(4) 457ndash 471

ISO (1997) The Virtual Reality Modeling Language Inter-national Standard ISOIEC 14772-11997 [on-line] Avail-able httpwwwvrmlorgSpecicationsVRML97

Kjellstrom B (1976) Be Expert with Map and Compass TheComplete Orienteering Handbook New York Scribner Pub-lishers

Liggett R Friedman S amp Jepson W (1995) Interactivedesigndecision making in a virtual urban world visual sim-ulation and GIS Proceedings of the 1995 ESRI User Confer-ence Environmental Systems Research [on-line] Availablehttpwwwesricombasecommonuserconfproc95

Lynch K (1960) The Image of the City Cambridge MAMIT Press

Elvins et al 581

landmark in the city We observed that some subjectsoften forgot what it was they were looking for and sohad to open the guidebook momentarily to remindthemselves

Not surprisingly landmarks with prominent signage(such as ldquoBobrsquos Groceryrdquo) and landmarks painted withmore highly saturated colors seemed easier to nd thanother types of landmarks Also after passing all thelandmarks several subjects exclaimed that they had lostall sense of direction within the world

A nal unexpected result were the many similarities be-tween the text and image guidebooks The text and imageguidebooks were only signicantly different in the timesubjects spent consulting them Considering that text andimages represent landmarks in very different ways we ex-pected differences in time and distance measures

76 Major Findings

In interpreting these ndings it is important toremember that the subjects performed the wayndingtasks using only the landmark knowledge gained bystudying the guidebooks No procedural or survey(map) information was provided

c As we predicted the total time spent traveling tothe goal was signicantly less using the worldletguidebook than when using the text or imageguidebooks

c Subjects spent signicantly more time studying theworldlet guidebook than they did the text or imageguidebook

c Increased worldlet study time was more than com-pensated for by a reduction in the time required toreach the goal Using the worldlet guidebook sub-jects traveled an average of 55 faster than whenusing the text guidebook and 38 faster than whenusing the image guidebook

c Increased worldlet study time also allowed subjectsto move more directly to the goal and reducedbacktracking time to almost zero This is showndiagramatically in gure 13 in which gures 13aand 13b show brighter wider streets than in gure

13c Figure 13c also exhibits fewer detours and lesswandering than do gures 13a and 13b

We postulate that compared to the landmark knowl-edge gained from text and image guidebooks the im-proved landmark knowledge built while consulting theworldlet guidebook enabled subjects to refer less fre-quently to the guidebook This postulation is furthersupported by examining worldlet guidebook results

c Subjects spent signicantly more time studying theworldlet guidebook than they did the other types ofguidebooks

c When using the worldlet guidebook subjects trav-eled faster more directly and with more skill tothe goal kiosk

c Subjectsrsquo median rating for the difculty of thesearch task was Easy when using the worldlet guide-book

8 Conclusion

The waynding literature shows that landmarksare critical in navigating an environment whether it isreal or virtual While building landmark knowledgetravelers characterize a landmark representation by its3-D shape size texture context and so forth In awaynding task textual or image guidebook landmarkdescriptions fail to engage the full range of 3-D land-mark characteristics recognized and used by travelers tond their way

We know of no studies that have considered the com-parative advantages of different types of guidebooks forwaynding Intuitively one would predict that 3-Dthumbnails make better guidebooks for 3-D worldsthan do text and 2-D thumbnail images and indeedthis is what we found We wanted however to identifythe component activities in using guidebooks that mostbenetted from 3-D elements Perhaps of greatest inter-est was our nding that worldlets enabled subjects tond their way in complex environments with almost nobacktracking By spending more time studying theworldlets in their guidebooks they spent less time look-ing around reorienting themselves conducting exhaus-

580 PRESENCE VOLUME 10 NUMBER 6

tive searches and performing similar time-wasting activ-ities We must be cautious however in what weconclude because worldlets contain more informationthan do text or 2-D thumbnail images To control forthis difference in information content more experi-ments are necessary

9 Future Work

We would like to investigate the comparative ad-vantages of worldlet guidebooks for waynding tasks incorresponding virtual and real world environments(both urban and rural) We are also considering an ex-periment to evaluate tandem use of traditional 2-Dmaps and landmark guidebooks Requiring effectivemap and compass use in a natural environment thesport of orienteering (Kjellstrom 1976) might benetfrom worldlets By encapsulating the region around theagged control points into a worldlet orienteers couldgain rst-person experience with natural landmarks suchas foliage rocks gullies shadows and so on beforebeginning the search

Acknowledgments

Our appreciation goes to Planet 9 Studios (wwwplanet9com)for the use of their city building textures Thanks to JohnMoreland and Mike Bailey for technical assistance on thisproject and thanks to Suzi Feeney for collaborating in thepilot study design UCSD librarian Debbie Cox and her staffwere invaluable in locating references Thanks also to JohnHelly and Reagan Moore for their support San Diego Super-computer Center is funded by NSF the University of Califor-nia the State of California and industrial partners

References

Appleyard D A (1976) Planning a Pluralistic City Con-icting Realities in Ciudad Guayana Cambridge MAMIT Press

Art1 Com (1997) The cybercity Berlin project [on-line]Available httpwwwartcomdeprojects

Chen S E (1995) QuickTime VRmdashAn image-based ap-proach to VE navigation Proceedings of SIGGRAPH 9529 ndash38

Darken R P amp Sibert J L (1996) Waynding strategiesand behaviors in large virtual worlds Proceedings of CHI 96142ndash149

Downs R J amp Stea D (1973) Cognitive maps and spatialbehavior In R J Downs amp D Stea (Eds) Image and En-vironment Chicago Aldine Publishing Company 8 ndash26

Elvins T T (1997) Waynding in layered informationworlds Proceedings of the CODATA Euro-American Work-shop on Data and Information Visualization 6 ndash7

Elvins T T Nadeau D R amp Kirsh D (1997) World-letsmdash3D thumbnails for waynding in virtual environmentsProceedings of the ACM User Interface Software and Technol-ogy Symposium 21ndash30

Elvins T T (1998) WorldletsmdashA novel 3D thumbnail af-fordance for waynding in large virtual environmentsUnpublished doctoral dissertation Department of Elec-trical and Computer Engineering University of Califor-niandashSan Diego

Elvins T T Nadeau D R Kirsh D amp Schul R (1998)Worldletsmdash3D thumbnails for 3D browsing Proceedings ofthe ACM CHI Conference 21ndash30

Foley J van Dam A Feiner S amp Hughes J (1990) Com-puter Graphics Principles and Practice Reading MA Addi-son-Wesley

Gillner S amp Mallot H A (1998) Navigation and acquisi-tion of spatial knowledge in a virtual maze Journal of Cog-nitive Neuroscience 10(4) 445ndash 463

Goldin S E amp Thorndyke P W (1982) Simulating naviga-tion for spatial knowledge acquisition Human Factors24(4) 457ndash 471

ISO (1997) The Virtual Reality Modeling Language Inter-national Standard ISOIEC 14772-11997 [on-line] Avail-able httpwwwvrmlorgSpecicationsVRML97

Kjellstrom B (1976) Be Expert with Map and Compass TheComplete Orienteering Handbook New York Scribner Pub-lishers

Liggett R Friedman S amp Jepson W (1995) Interactivedesigndecision making in a virtual urban world visual sim-ulation and GIS Proceedings of the 1995 ESRI User Confer-ence Environmental Systems Research [on-line] Availablehttpwwwesricombasecommonuserconfproc95

Lynch K (1960) The Image of the City Cambridge MAMIT Press

Elvins et al 581

tive searches and performing similar time-wasting activ-ities We must be cautious however in what weconclude because worldlets contain more informationthan do text or 2-D thumbnail images To control forthis difference in information content more experi-ments are necessary

9 Future Work

We would like to investigate the comparative ad-vantages of worldlet guidebooks for waynding tasks incorresponding virtual and real world environments(both urban and rural) We are also considering an ex-periment to evaluate tandem use of traditional 2-Dmaps and landmark guidebooks Requiring effectivemap and compass use in a natural environment thesport of orienteering (Kjellstrom 1976) might benetfrom worldlets By encapsulating the region around theagged control points into a worldlet orienteers couldgain rst-person experience with natural landmarks suchas foliage rocks gullies shadows and so on beforebeginning the search

Acknowledgments

Our appreciation goes to Planet 9 Studios (wwwplanet9com)for the use of their city building textures Thanks to JohnMoreland and Mike Bailey for technical assistance on thisproject and thanks to Suzi Feeney for collaborating in thepilot study design UCSD librarian Debbie Cox and her staffwere invaluable in locating references Thanks also to JohnHelly and Reagan Moore for their support San Diego Super-computer Center is funded by NSF the University of Califor-nia the State of California and industrial partners

References

Appleyard D A (1976) Planning a Pluralistic City Con-icting Realities in Ciudad Guayana Cambridge MAMIT Press

Art1 Com (1997) The cybercity Berlin project [on-line]Available httpwwwartcomdeprojects

Chen S E (1995) QuickTime VRmdashAn image-based ap-proach to VE navigation Proceedings of SIGGRAPH 9529 ndash38

Darken R P amp Sibert J L (1996) Waynding strategiesand behaviors in large virtual worlds Proceedings of CHI 96142ndash149

Downs R J amp Stea D (1973) Cognitive maps and spatialbehavior In R J Downs amp D Stea (Eds) Image and En-vironment Chicago Aldine Publishing Company 8 ndash26

Elvins T T (1997) Waynding in layered informationworlds Proceedings of the CODATA Euro-American Work-shop on Data and Information Visualization 6 ndash7

Elvins T T Nadeau D R amp Kirsh D (1997) World-letsmdash3D thumbnails for waynding in virtual environmentsProceedings of the ACM User Interface Software and Technol-ogy Symposium 21ndash30

Elvins T T (1998) WorldletsmdashA novel 3D thumbnail af-fordance for waynding in large virtual environmentsUnpublished doctoral dissertation Department of Elec-trical and Computer Engineering University of Califor-niandashSan Diego

Elvins T T Nadeau D R Kirsh D amp Schul R (1998)Worldletsmdash3D thumbnails for 3D browsing Proceedings ofthe ACM CHI Conference 21ndash30

Foley J van Dam A Feiner S amp Hughes J (1990) Com-puter Graphics Principles and Practice Reading MA Addi-son-Wesley

Gillner S amp Mallot H A (1998) Navigation and acquisi-tion of spatial knowledge in a virtual maze Journal of Cog-nitive Neuroscience 10(4) 445ndash 463

Goldin S E amp Thorndyke P W (1982) Simulating naviga-tion for spatial knowledge acquisition Human Factors24(4) 457ndash 471

ISO (1997) The Virtual Reality Modeling Language Inter-national Standard ISOIEC 14772-11997 [on-line] Avail-able httpwwwvrmlorgSpecicationsVRML97

Kjellstrom B (1976) Be Expert with Map and Compass TheComplete Orienteering Handbook New York Scribner Pub-lishers

Liggett R Friedman S amp Jepson W (1995) Interactivedesigndecision making in a virtual urban world visual sim-ulation and GIS Proceedings of the 1995 ESRI User Confer-ence Environmental Systems Research [on-line] Availablehttpwwwesricombasecommonuserconfproc95

Lynch K (1960) The Image of the City Cambridge MAMIT Press

Elvins et al 581

Passini R (1992) Waynding in Architecture (2nd ed) NewYork Van Nostrand Reinhold

Pausch R Burnette T Brockway D amp Weiblen M E(1995) Navigation and locomotion in virtual worlds viaight into hand-held miniatures Proceedings of SIGGRAPH95 399 ndash400

Peponis J Zimring C amp Choi Y K (1990) Finding thebuilding in waynding Environment and Behavior 22(5)555ndash590

Ruddle R A Payne S J amp Jones D M (1998) Navigat-ing large-scale desk-top virtual buildings Presence Teleop-erators and Virtual Environments 7(2) 179 ndash192

Satalich G A (1995) Navigation and Waynding in Vir-tual Reality Finding the Proper Tools and Cues to En-hance Navigational Awareness Unpublished mastersthesis Department of Computer Science University ofWashington [on-line] Available httpwwwhitlwashingtonedupublications

University College London Center for Advanced Spatial Anal-ysis Virtual London A proposal [on-line] Availablehttpwwwcasauclacuk30dayslondonhtm

Waller D Hunt E amp Knapp D (1998) The transfer ofspatial knowledge in VE training Presence Teleoperators andVirtual Environments 7(2) 129 ndash143

Wechsler D (1955) Manual for the Wechsler Adult Intelli-gence Scale New York Psychological Corporation

Wernecke J (1994) The Inventor Mentor Reading MA Ad-dison Wesley

Wickens C amp Baker P (1995) Cognitive issues in virtualreality In Woodrow Bareld and Thomas Furness III(Eds) Virtual Environments and Advanced InterfaceDesign (pp 514 ndash541) New York Oxford UniversityPress

Witmer G G Bailey J H Knerr B W amp Parsons K C(1996) Virtual spaces and real world places Transfer of routeknowledge Int J Human-computer Studies 45 413ndash428

582 PRESENCE VOLUME 10 NUMBER 6