computation & crowds: models for dynamic ridesharing · computation & crowds: models for...
TRANSCRIPT
![Page 1: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/1.jpg)
Computation & Crowds:
Models for Dynamic Ridesharing
Article:
Collaboration and Shared Plans in the Open World: Studies of Ridesharing,
IJCAI 2009, Pasadena, CA, July 2009.
Ece Kamar and Eric Horvitz
Microsoft Research
![Page 2: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/2.jpg)
Computation at center of opportunistic
planning & coordination among people
Principles for weaving together collaborations
in light of related goals and preferences
Rideshare as theoretical & practical example
![Page 3: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/3.jpg)
with John Krumm
Multiple sources
- GPS, cell tower, wifi
- Direction requests to routing services
> e.g., MS Multiperson Location Survey
![Page 4: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/4.jpg)
with John Krumm
Multiple sources
- GPS, cell tower, wifi
- Direction requests to routing services
> e.g., MS Multiperson Location Survey
![Page 5: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/5.jpg)
with John Krumm
e.g. Predestination algorithm
Trip starts Inference as trip
progresses
Narrowing of destination
possibilities
Assign probabilities to each cell as trip progresses
• Median error 2 kilometers at trip’s halfway point
![Page 6: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/6.jpg)
5 yrs of GPS trails
~500,000 km
Multiple projects- Clearflow (now in 72 cities)
- Community sensing
![Page 7: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/7.jpg)
e.g., Extract AM/PM commutes to/from Microsoft
![Page 8: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/8.jpg)
Ongoing computation in support of
collaboration
Changing needs & preferences
Acceptance, trust, convenience, cost
Range of scenarios
Spectrum across immediacy vs. planned
General vs. special situation
Owned car vs. shared vehicle (e.g., Zipcar style)
![Page 9: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/9.jpg)
Cost-benefit
Earlier departure
Delayed arrival
Increased travel
Savings on effort
Fuel, environment
Start time
to
+-
Start time
to
+-
Arrive time
to
+-
Arrive time
to
+-
D Trip Duration
to
+-
D Trip Duration
to
+-
![Page 10: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/10.jpg)
Arrive time
to
+-
Arrive time
to
+-
Cost-benefit
Earlier departure
Delayed arrival
Increased travel
Savings on effort
Fuel, environment
Start time
to
+-
Start time
to
+-
D Trip Duration
to
+-
D Trip Duration
to
+-
![Page 11: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/11.jpg)
Instant & planned rideshare scenarios
Methods for promoting fairness in reporting needs
Social relationships, comfort, communication
Prototype for running system & analytical bench
Optimize for individuals and across a population
![Page 12: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/12.jpg)
Cost-benefit methods to find ideal rideshares
Evaluated on GPS trails from MS employees
with Ece Kamar
![Page 13: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/13.jpg)
Assignments based on observed trips.
Cost-benefit
- Departure change
- Delayed arrival
- Increased travel
- Savings on effort, fuel, environment
![Page 14: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/14.jpg)
![Page 15: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/15.jpg)
Planned commute
ABC notified of AM/PM needs day in advance
Instant: Commute requests on the fly
ABC notified 15 minutes before trip start time
![Page 16: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/16.jpg)
![Page 17: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/17.jpg)
New user
(<30min)
Original
Route
![Page 18: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/18.jpg)
Rideshare
Cost-benefit
Analysis
Queue
![Page 19: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/19.jpg)
Trip Activity
-Green: share
-Red: single
Queue
![Page 20: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/20.jpg)
![Page 21: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/21.jpg)
0
10
20
30
40
50
60
108 215 430 860
Syst
em
Eff
icie
ncy
Number of agents
Efficiency on number of commutes
Efficiency on total cost
Number of participants
![Page 22: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/22.jpg)
Fuel Cost
![Page 23: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/23.jpg)
Cost of time
![Page 24: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/24.jpg)
Challenge: Understanding acceptance,
perceptions, social considerationsAddress concerns, leverage opportunities
Trusted organizations
Referral, reputation
- e.g., existing online social networks (e.g., link distance bounds)
?
![Page 25: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/25.jpg)
- Constraints
• - Preferences
= f(d(ai1,aj1),…, d(ain,ajn))
= Sl kld(ail,ajl)
U(pi,pj) = d(ai,aj)
Optimization allows for smooth insertion of:
![Page 26: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/26.jpg)
![Page 27: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/27.jpg)
![Page 28: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/28.jpg)
Studies of preferences & acceptability
Flexibility, acceptance, and ease of use
Implementation directions Shuttle overlay, instant carpool, AM/PM commute
Outlook add-in, web service
- Encode preferences, needs, commitments
Collaboration with MS Real Estate & Facilities,
MS Sustainability, King County Metro
![Page 29: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/29.jpg)
Identified methods that harness computing
for coordination among people
Computing services integrate data about
know-how, availability, location, ad goals.
Other location-centric coordination
![Page 30: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/30.jpg)
e.g., Related work: Creating meshes of people
for emergency assistance
with Jure Leskovec
![Page 31: Computation & Crowds: Models for Dynamic Ridesharing · Computation & Crowds: Models for Dynamic Ridesharing Article: Collaboration and Shared Plans in the Open World: Studies of](https://reader035.vdocuments.us/reader035/viewer/2022062917/5ed37bf2847f87317f77beee/html5/thumbnails/31.jpg)