These are research notes for the travel modeling project.

People

• Alan Borning
• Dieter Fox
• Paul Wadell
• Peter Henry
• Alice Neels

Steps To Take

Paul's steps:

1. Route choice for Home -> Work trip (known start point and destination, possibly incorporate traffic info)
   • How do we anticipate traffic volumes on side streets? (Highway have loop sensors)
2. Dynamic congestions modeling?
   • Simulate all of Seattle (or something) - model that converges to actual observed traffic patterns
   • traffic microsimulation?
   • queuing technique
   • car following technique (these two are different)
   • would involved learning distributions on departure/arrival times and constraints/elasticity
   • also need to learn mode and route distributions
3. use this to estimate how people value time (based on salary, circumstances, etc.)
   • an interesting use of toll data
4. question of intermediate stops (along a commuting route or such)
   • pick up the kids, go shopping
   • more complex route planning
   • can't do this with cab data
   • first step to hierarchical learning
   • primary and secondary tours (get lunch from work...)
   • timespace prism
5. overall activity planning for a family
   • incorporate traffic information
   • how people value different kinds of time (times spent at home, traveling, at work, etc.)
   • constraints on trace through timespace prism (costs of being late etc.)
   • mandatory vs. discretionary activities

Other stuff:

1. (Peter) snap GPS traces
   • need to NumPyify? this stuff
2. Implement our own version of Ziebart, Bagnell et al paper
   • implement in Python with NumPy/SciPy? for computation
3. Apply this to PSRC data
4. Look at commute planning
5. Use this plus panel survey data to look at overall family activity planning

This section needs to be supplemented with Paul's notes from 4/28 or another discussion on the topic. How does congestion pricing factor into this? When and how could this translate into an UrbanSim module for traffic simulation?

Data

• PSRC has good road data for roads mapped to GPS grids. We have a copy of this already (talk to Jesse). It has feature data (speed limit, etc.) for each street segment. However, small local streets aren't all included.
• Does Yahoo or Google have this data somewhere?

PSRC

The PSRC traffic choices study collected GPS traces from vehicles driven by families in the Seattle area. We hope to someday get our hands on this data. Their reports page has the results of their study of this data. It appears that most of their analysis was done on totals of trips, miles driven, etc. The elasticities of these quantities under pricing changes were measured. Changes in travel times (i.e. when someone left for work) were also examined.

Related Work

• Ziebart, B., et al. Maximum Entropy Inverse Reinforcement Learning (sent out by Dieter, will be in AAAI?)
• Abeel paper