nextgenintelligenttransportaon: measuring’people ... · • in partnership with sfmta • 6 weeks...
TRANSCRIPT
NextGen Intelligent Transporta2on: Measuring People, Controlling Things
Raja Sengupta, Professor Systems, CEE, UC Berkeley [email protected],
5107170632
Berkeley Terraswarm project
• CENTER MISSION: The TerraSwarm Research Center (TSRC) aims to enable the simple, reliable, and secure deployment of advanced distributed sense-‐control-‐actuate applicaSons on shared, massively distributed, heterogeneous, and mostly uncoordinated swarm plaTorms through an open and universal systems architecture.
• In normal operaSon, a swarm-‐enabled \Smart City" not only helps run the infrastructure more eecSvely but empowers its occupants by providing more eecSve interfaces, beWer mobility, and experiences in immersive realiSes in a way not possible before. For example, maintenance crews may recruit sensors from underground uSliSes, and combine that sensor data with data from pipe-‐crawling robots and from the cloud. They can use this informaSon to guide maintenance operaSons using overlay displays in a manner similar to what televised sporSng events use, based on contextual 3D informaSon.
Human and Social Capital
• Smart CiSes , Caragliu etal 2009 – Urban performance currently depends not only on the city's endowment of hard infrastructure ('physical capital'), but also, and increasingly so, on the availability and quality of knowledge communicaSon and social infrastructure ('intellectual capital and social capital'). …… It is against this background that the concept of the smart city has been introduced as a strategic device to encompass modern urban producSon factors in a common framework and to highlight the growing importance of InformaSon and CommunicaSon Technologies (ICTs), social and environmental capital ……
The benefits of a city investment are determined by the produc2on func2on of the investment and the use its
ci2zens make of it.
• In transportaSon ciSzens control – AcSvity choice – Trip chaining – Trip choice – Mode choice – Departure Sme choice
– Route choice
The inter-‐play of the physical ar2fact and its use can be important
• In transportaSon ciSzens control – AcSvity choice – Trip chaining – Trip choice – Mode choice – Departure Sme choice
– Route choice • From
hWp://thestandinginvitaSon.wordpress.com/tag/game-‐theory/
The User Equilibrium is not System Op2mal
• In transportaSon ciSzens control – AcSvity choice – Trip chaining – Trip choice – Mode choice – Departure Sme choice – Route choice
• Making the inter-‐play of the two important
• From hWp://thestandinginvitaSon.wordpress.com/tag/game-‐theory/
It’s 2002. You are the mayor of Seoul. There is a major traffic crisis in the city, with congesSon rising by as much as 5% yearly. You have £200 million to spend on solving the problem. What do you do? Surely you build more roads to ease the traffic… right? Surprisingly, the mayor did the opposite: he spent the money demolishing roads. Even more surprisingly, it worked. Because of the strange fact of Braess’s Paradox, shuing down roads can actually decrease traffic. Here’s how.
To realize the benefits of an investment one must Adapt the City to the Ci2zen and The Ci2zen to the City
Feedback Control Theory is one science of Adapta2on
ITS is older on the Supply-‐side More recent on Demand
• For safety: Collision Warning, Lane Keeping,… • For Increasing Supply
– Ramp Metering – Coordinated Arterial Signal Control – Bus fleet management systems – Electronic Toll CollecSon
• For Managing Demand (CiSzen Choices) – Cordon Pricing/CongesSon Pricing – Dynamic Parking Pricing – Spare the Air Day, Bike to Work Day – Travel Feedback Programs
ITS is older on the Supply-‐side More recent on Demand
• For safety: Collision Warning, Lane Keeping,… • For increasing supply
– Ramp Metering – Coordinated Arterial Signal Control – Bus fleet management systems – Electronic Toll CollecSon
• For Managing Demand/User Choice – Cordon Pricing/CongesSon Pricing – Dynamic Parking Pricing – Spare the Air Day, Bike to Work Day – Travel Feedback Programs
Done as Feedback control
These are NOT
The difference between traffic management and demand management is the difference between control and planning
Survey Op2mize Intervene
Deploy
Parking price Cordon price Spare the air day
Link speed measurement
Link inflow (op2mal) control
2me
2me
Ramp Metering
Demand Management
Evaluate Baseline
ITS is older on the Supply-‐side More recent on Demand
• For safety: Collision Warning, Lane Keeping,… • For increasing supply
– Ramp Metering – Coordinated Arterial Signal Control – Bus fleet management systems – Electronic Toll CollecSon
• Managing Demand/AdapSng User Choice – AcSvity choice – Trip chaining – Trip choice – Mode choice – Departure Sme choice – Route choice
Done as Feedback control
NOT done as feedback control Survey à OpSmize à Intervene à Evaluate
Why?
The SF Travel Quality Study • In partnership with SFMTA • 6 weeks • 800+ participants • Entry, exit survey • Location-aware survey app
• Take Muni, use our app How do you feel? How was the ride? – 13,000 responses
• Real-time subject location for 6 weeks – 82 million location points
• Real-time transit vehicle location
Mobile survey • How did this Muni
experience make you feel?
– Pleased – Frustrated – Relaxed – Impatient
• How do you generally feel today?
• How satisfied were you with…
– Reliability – In-vehicle travel
time – Wait time – Transfer time – Accuracy of
real-time information
• How satisfied were you with…
– Cleanliness – Crowding – Pleasantness of
other passengers
– Safety
EmoSons OperaSons Environment
Matching the data Smartphone locaSon data
Transit vehicle raw AVL data
Transit runs
Personal reliability metrics Route shape
files
Personal transit
travel diary
Timetable data
SaSsfacSon surveys
2
1
3
4
• Strong effect of in-vehicle delays on satisfaction and choice behavior
How do surveys relate to engineering?
• Responses differ between online surveys and mobile surveys (Question presentation? Environment?)
Survey: Presentation, Context?
Daily mobile survey, graphical labels L -‐ J
End of study mobile survey, graphical labels L -‐ J
End of study online survey, text (Likert) labels
Quan2fied Traveler Part 1: Automated Travel Diaries
ICTBR 2012, Jariyasunant Phd thesis UCB
Phone App: GPS, Accelerometer, WiFi, Cell Tower data + Cloud Server: Processing it, serving web pages
The Automated Trip Diary
• No User Input – Automated Signal Processing and Machine Learning for Trip origin,
des2na2on, 2me, route, mode • Learns HotSpots and HotRoutes • Learns HotModes with user input
Quantified Traveler
An Evaluation • 3 weeks March 18 to April 7, 2012 • Pre and Post Experiment Survey
– 55 statements on awareness, experiential and instrumental attitudes, moral obligations, perceived norms, perceived control, self efficacy
• Installed the QT App on personal Iphones and Android phones
• Week 1 no feedback, 7th day given link to QT, Week 2 Spring Break data, Week 3 feedback – Week 2 data not used in behavior change analysis
An Evaluation • 135 subjects
– No one uninstalled the App, 118 completed both surveys – 4143 trips logged on 5 modes (bike, walk, drive, bus, train) – 258 km per subject on average – Looked at website 4.1 times on average in week 2 – Participants corrected mode on 13.5% of trips
• 78 subjects used for behavior change analysis
Users stop using the average applications quickly. Long term audiences are generally 1% of total downloads" - Pinch Media http://www.techcrunch.com/2009/02/19/pinch-media-data-shows-the-average-shelf-life-of-an-iphone-app-is-less-than-30-days
An Evaluation • 135 subjects
– No one uninstalled the App, 118 completed both surveys – 4143 trips logged on 5 modes (bike, walk, drive, bus, train) – 258 km per subject on average – Looked at website 4.1 times on average in week 2 – Participants corrected mode on 13.5% of trips
• 78 subjects used for behavior change analysis
• Basic Finding: Behavior change from week 1 to 2 – Decrease in driving: Frequent drivers 38% reduction
(p<0.01), Infrequent drivers 27% (p=0.09) – Increase in walking (p-value 0.03)
Travel Feedback Programs: Travelers meet with a counselor and receive personalized advice
High rates of success switching travelers to more sustainable modes
QT as a con2nuous feedback System
Interven2on
Deploy
2me
Evaluate Baseline
Deploy
Sense
QT is a con2nuous feedback system
• Its design theory is in Psychology. Not a control theory.
Daily travel
SmartPhone Travel Diary
QT Web Server
Time Cost Exercise Green
Traveler Choices
People persuade people. Can computa2onal systems persuade people?
Personalized feedback
Personalized feedback ?
Successful Actuation needs Aesthetics and good HCI
• Summary page liked the most followed by breakdown and trip diary. Liked comparison to others in the group. – 4 questions on 7-point Likert scale about the website.
• I enjoyed taking a look at my dashboard/statistics/trip history page and getting a summary of my travel – Mean = 5.4, Std Dev = 1.2
• In the future, this web page is something I would consider using. – Mean = 5.9, Std Dev = 1.0
• If I were to set a goal to change my travel behavior (be greener, reduce cost, travel less), I consider this web page helpful. – Mean = 5.1, Std Dev = 1.3
• This web page was easy to use. Mean = 5.3, Std Dev = 1.1
Joint work with
• Andre Carrel, Venky Ekambaram, DJ Gaker, Dr. Jerry Jariyasunant, Thejo Kote, Eric Mai
• Several undergraduate students • Prof. Joan Walker, Prof Maya Abou Zeid • Spin-‐off Ventures 2010 –
– automaSc.com – Lockx.com – Baytripper.org – Human Intellect Lab (hWp://myne.net/)
Some ITS Theories and Systems
• For safety: Collision Warning, Lane Keeping,… • For increasing supply
– Ramp Metering – Coordinated Arterial Signal Control – Bus fleet management systems – Electronic Toll CollecSon
• Managing Demand/AdapSng User Choice – AcSvity choice – Trip chaining – Trip choice – Mode choice – Departure Sme choice – Route choice
Done as Feedback control
This is NOT
Why?
Because the technological arSfacts for the sensing and actuaSon of user choice did not exist unSl recently On the supply-‐side they have existed for almost two decades
Rise of Ubiquitous Computing: Continuous Sensing and Actuation for User Choice
• Quantified Traveler is a part of the Quantified Self genre
-‐ Record Finances -‐ Set Goals -‐ Analyze Personal Spending -‐ Analyze Trends
Behavior Change or Persuasive Technology
• Quantified Self + • Closing the loop with an objective
– With the consent of the subject – Better eating, Better exercise, Sustainable travel – Ubifit, Fish N Steps, Healthy Lifestyle Coach, Designing Games to Mo2vate Physical Ac2vity, Persuasive Picture Frames for Proper Posture,
– Persuasive Conference
Sensing and ActuaSon Has Arrived for Behavioral Systems
• For CiSzen Choices – Cordon Pricing/CongesSon Pricing – Dynamic Parking Pricing – Spare the Air Day, Bike to Work Day – Travel Feedback Programs
Sensing smartPhone Fitbit OFX NEST ….. Con2nuously, Longitudinal depth
Sensing and ActuaSon Has Arrived for Behavioral Systems
• For CiSzen Choices – Cordon Pricing/CongesSon Pricing
– Dynamic Parking Pricing – Spare the Air Day, Bike to Work Day
– Travel Feedback Programs
And ActuaSon Apps Tweets Facebook Recommenda2on engines…. Con2nuously, Longitudinal depth
Pozdnoukhov and Coffey
Bikeshare Commute Data
TwiWer Topic Analysis
Granger Causality Analysis
Behavioral Systems as Dynamical Systems What part is generalizable?
The Berkeley Approach PlaTorms, Colleagues, ExploraSons
Psychology, Economics, and ComputaSon
Shachar Kariv, Economics
Scob Moura, Control, Energy
Alexei Pozdnoukhov, GIS Computa2onal Sociology
Elizabeth Deakin, City Planning
Joan Walker Global Metropolitan Studies
John Canny HCI, Behavioral Data Mining
Raja Sengupta Control Theory
Design Theory in Psychology QT rests on the Theory of Planned Behavior (Azjen 1991)
• Psychological variables: – Awareness of travel impacts is associated with more favorable
aitudes towards sustainable transportaSon behavior – Peer influence effect is significant:
• Favorable comparison of travel impacts to one’s peers leads to more posiSve aitudes
• Unfavorable comparison has the opposite effect
• Behavior change: – More posiSve aitudes and norms are associated with a significant
reducSon in driving – Regular drivers decrease their driving most, while students decrease
walking/biking distance – More frequent feedback (logins to the website) is associated with an
increase in walking/biking distance
Change in walk /bike distance
Change in adtude
Change in awareness
Change in norms
Change in distance driven
Peer influence
Driver Student
Logins
Indicators
Indicators
Indicators
The QT Structural EquaSon Model
Change in walk /bike distance
Change in adtude
Change in awareness
Change in norms
Change in distance driven
Peer influence
Driver Student
Logins
Indicators
Indicators
Indicators
The QT Structural EquaSon Model
0.102 (0.66)
0.180 (1.35)
0.0352 (2.51)
-‐0.611 (-‐2.70)
0.00755 (0.43)
0.0117 (0.93)
-‐0.224 (-‐2.16)
0.0125 (2.01)
-‐0.121 (-‐3.94)
-‐0.737 (-‐3.35)
0.0433 (2.99)
GeneralizaSon is done by demographics: Age, Income, …
For PersonalizaSon: There is an incompleteness problem
Separate into Es2ma2on and Control Experimental Micro-‐Economics à Func2onal Forms for the Es2ma2on
Problem • Four fundamental tradeoffs
– Risk vs. Return – Today vs. Tomorrow – Work vs. Leisure – Self vs. Others
• ComputaSonal toolkits to reveal parameters (preferences) -‐ Recoverability
$100
$40 $30
$30
Risk vs. Return How do you choose?
Heads
Tails
Kariv 2007
EsSmate parameters for Gul’s model loss and disappointment aversion With CARA specificaSon
The Laboratory is in the Wild
• Four fundamental tradeoffs – Risk vs. Return – Today vs. Tomorrow – Work vs. Leisure – Self vs. Others
• ComputaSonal toolkits to reveal preferences
$100
$40 $30
$30
Risk vs. Return How do you choose?
Heads
Tails
Kariv 2007
EsSmate parameters for Gul’s model loss and disappointment aversion With CARA specificaSon
Android Screen
Time of Day Effect Day of Week Effect Effect of Sme of day on percentage allocated to the cheapest asset. ID 73261: ID 74188:
Effect of day of the week on percentage allocated to the cheapest asset. ID 77848: ID 62896:
The dynamics of preference? Insufficient data
Sensing and ActuaSon Has Arrived for Behavioral Systems
• For CiSzen Choices – Cordon Pricing/CongesSon Pricing – Dynamic Parking Pricing – Spare the Air Day, Bike to Work Day – Travel Feedback Programs
Sensing smartPhone Fitbit OFX NEST ….. Con2nuously, Longitudinal depth
Sensing and ActuaSon Has Arrived for Behavioral Systems
• For CiSzen Choices – Cordon Pricing/CongesSon Pricing
– Dynamic Parking Pricing – Spare the Air Day, Bike to Work Day
– Travel Feedback Programs
And ActuaSon Apps Tweets Facebook Recommenda2on engines…. Con2nuously, Longitudinal depth