Uncertainty in Trip Planning in Transportation Systems A. P. Sistla (Joint work with Booth, Wolfson and Cruz) Features of Urban Transportation Multi-modal: Trains, buses, auto, pedestrians... Uncertainties in travel times Provides facilities at different stops Graph Model of the network Use labeled Multi-graph Nodes --- Represent stops and stations Edges --- Connections by different modes Node Labels --- Name, Geometry and Facilities Facilities : restaurants, supermarkets, etc. Edge Labels--- Mode, Route#, Run#, Departure time, Duration Example Network Querying Trips Find a trip from work to home leaving after 5PM and reaching before 6PM. Trip may involve mutiple modes, require some stopping points. Some Definitions: LEG: A path with all edges having same mode, route#, run#. TRIP: A sequence of legs (L1,L2,...,Ln) end point of each leg is the start if the next leg. Trip Example Query Language Should be easy to use. Should be able to specify intermediate stops, Certainty as probability Conditions involving on travel times Use an operator ALL-TRIPS(Origin, Destination) Employ SQL-like syntax Query Format SELECT * FROM ALL-TRIPS (origin,destination) WITH Stop-vertices WITH Modes WITH CERTAINTY prob-value WHERE condition OPTIMIZE ttribute Example SELECT * FROM ALL-TRIPS(work, home) as t WITH MODES pedestrian, bus WITH CERTAINTY 0.8 WHERE finish-time(t) <= 5PM CERTAINTY clause specifies the probability that the trip satisfies the where condition Another Example SELECT * FROM ALL-TRIPS(home, theater) as t WITH STOP POINT v WHERE “ATM” in v.facilities AND start-time(t) >= 2PM AND finish-time(t) <=3PM WITH CERTAINTY 0.95 Semantics and Processing of Queries Need precise semantics of queries including the certainty clause. Consider the “home-to-theater” query Let F be the set of possible trips Consider a trip t = (L1,L2) from F Let v be the intermediate stop. Evaluate and simplify the where condition on trip t. Assume it satisfies non-temporal conditions. Let C be the resulting condition. Query Processing Continued In the example C is --- start-time(t)>= 2PM AND finish-time(t)<=3PM. Let Y1, Y2 be random variables denoting departure times of legs L1 and L2. Let Z1, Z2 be the random variables representing durations of L1 and L2. We want to compute the probability that trip t satisfies condition C. Computation of Probabilities We need joint density function of Y1,Y2,Z1,Z2. Let f(y1,y2,z1,z2) be this function. Departure time of L1 is Y1. Arrival time at destination is (Y2+Z2). Transform the condition C using y1,y2,z1,z2: y1 >= 2PM AND y2+z2 <= 3PM. Also need to make sure transfer at v is successful. Successful transfer probability What is the probability that the transfer at v is successful? Required stop time for successful transfer: RT Arrival time of L1 --- Y1+Z1 Duration at v: Y2 – (Y1+Z1). Transfer condition: (y2- (y1+z1)) >= RT Final Probability Combined condition: y1 >=2PM AND y2+z2 <= 3PM AND (y2-y1-z1) >=RT. The combined condition defines a region X in the 4-dimensional space. Probability t satisfies the where condition is the definite integral f(y1,y2,z1,z2) dy1 dy2 dz1 dz2 over the region X. Practical Approach Assume that travel times on different routes are independent. The joint density function can be written as product of independent density functions. Density functions can be maintained as tables. Further research needed for implementation. Alternates models of uncertainties Other possibilities: Using uncertainty intervals Other query constructs: Definitely and Possibly. Questions?