11 chapter 11: localization and map making a. occupancy grids b. evidential methods c. exploration...
TRANSCRIPT
11Chapter 11:
Localization and Map Making
a. Occupancy Gridsb. Evidential Methods
c. Exploration
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 2
11 Objectives
• Describe the difference between iconic and feature-based localization
• Be able to update an occupancy grid using either Bayesian, DS, or HIMM
• Describe the two types of formal exploration strategies
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 3
11 Navigation• Where am I going? Mission
planning
• What’s the best way there? Path planning
• Where have I been? Map making
• Where am I? Localization
MissionPlanner
Carto-grapher
BehaviorsBehaviorsBehaviorsBehaviors
deli
bera
tive
reac
tiveHow am I going to get
there?
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 4
11 Motivation• Can make topological or metric maps, localize relative
to landmark(s) or at any point• More desirable: metric maps, localize at any point
– More readable by a human
• GPS isn’t the answer– Localization error is on order of 1 meter– Reception difficult indoors– Want to know where features in environment are, not just
robot (e.g., layout of walls, not just robot’s path)
• Sensor measurements have some uncertainty that must be factored in– Formal methods called “evidential reasoning”, “theories of
evidence”
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 5
11 Basic Idea
• Sense and create a local map• Move a little
– Record change in position, orientation
• Sense and create a local map– Fuse/tile together
Localmap
Globalmap
Move
Integrate local map
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 6
11 Observations about Process• Map is almost always a type of regular grid (because easier to
visualize)
• The “Move ” and “Integrate local map” are the hard part. – Integration requires accurate measurement of (on order of inches
and <=5 degrees)
BlackIs groundTruth,Purple isMeasuredUsing shaftEncoders for
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 7
11 Iconic vs. Feature-Based
• Issue is how to localize at each step to accurately measure D, then integrate local map
• Iconic: use raw (or near raw) sensor readings– Match elements marked “empty” or “occupied” in a regular
grid• OCCUPANCY GRID
– Plug and chug, intense computations
• Feature-based: use features extracted from raw data– Label and match corners, walls, whatever
– Less features, so less computations
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 8
11 Occupancy Grids
• Type of regular grid– L: eLement
– Came out of sonar tradition
• Each element is marked with belief that L is empty or occupied– Usually a number on a
scale
– [0,1] for probability and possibility theories
– [0-15] for HIMM
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 9
11 Sonars and Occupancy Grids• Everything element L “under”
the sonar beam gets marked with some value for empty, occupied
• Exact value depends on– Sonar model
– Evidential method
• Generic sonar model– 3 regions
– R: theoretical range, r: measured range
: half angle
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 10
11 Evidential Methods for Occupancy Grids
• Bayesian– Popularized by Hans Moravec
• Dempster-Shafer
• HIMM– Johan Borenstein
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 11
11 Bayesian
• Compute the value for each L for each sonar using sonar model– The value of L is a probability
• Compute the value for each L where sonars overlap uses Bayes’ rule for updating
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 12
11 Example: Value of L in Region II
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 13
11 Class Exercise:Value of L in Region I
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 14
11 Other Issues
• An element L may have multiple “hits”– Robot moves and senses subset of same area, Sonars overlap:
what to do?
– Use Bayes’ rule to update
• If write a program to use Bayes’ rule, what’s the initialization of the occupancy grid?– P(Occupied)=P(Empty)=0.5
– Is this a good assumption?
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 15
11 Summary
• Localization and map making are intertwined– Localization requires good maps
– Map making requires good localization
• Map making and localization techniques often use occupancy grids– Type of regular grid
– Elements represent uncertainty of being empty, occupied
– Multiple ways of combining uncertainty when an element has multiple “hits”
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 16
11 Dempster-Shafer Theory & HIMM
• On board
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 17
11 Localization• Iconic: uses raw sensor data directly
– Ex. Sonar and laser readings fused in an occupancy grid– Compare current and past reading
• Feature-based: uses features extracted from sensor data– Ex. “corners”, “walls”
?
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 18
11 Iconic Example: ARIEL
• Issues– k must be small to be tractable, but k must be large if noisy sensors
– Doesn’t work with “just sonars”
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 19
11 Iconic Example: ARIEL
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 20
11 Results
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 21
11 Exploration
• Can explore reactively (move to open area as per Donath), but we’d like to create maps
• Two major methods– Frontier-based
– GVG
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 22
11 Frontier Based Exploration• Robot senses environment
• Borders of low certainty form frontiers
• Rate the frontiers– Centroid
– Utility of exploring (big? Close?)
• Move robot to the centroid and repeat
• (continuously localize and map as you go)
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 23
11 GVG
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 24
11 Keeps moving, ignores areas hard to get too
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 25
11 Reaches deadend at 9, backtracks
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 26
11 Goes back and catches missing areas
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 27
11 Discussion of Exploration
• Both methods work OK indoors, not so clear on utility outdoors
• GVG– Susceptible to noise, hard to recover nodes
• Frontier– Have to rate the frontiers so don’t trash
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary
Introduction to AI Robotics (MIT Press), copyright Robin Murphy 2000 Chapter 11: Localization and Map Making 28
11 Summary
• Map making requires– Localization and acurate measurements
– Exploration
• Localization and map making often use – Occupancy grids
– Evidential methods for updating• Bayesian
• DS
• HIMM (quasi-evidential)
• Two kinds of localization: iconic, feature-based
• Two popular methods for exploration: frontier-based, GVG
OverivewOccupancy Grids-Sonar Models-Bayesian Updating-Dempster-Shafer-HIMMLocalization-ARIELExploration-Frontier-based-GVGSummary