School of Civil Engineering

Learning Route Choice Behavior from Vehicle Trajectory Data for Traffic Origin-Destination Estimation

Conventional traffic measurements collected by loop detectors can capture all the vehicles passing over a detection zone, but not the entire road network. By contrast, new traffic data collected by Bluetooth scanners can capture the spatiotemporal trajectories of vehicles that provide a more comprehensive view of network flows. Yet only part of the vehicles can be detected due to low market penetration rates of such devices. This project aims to combine both aforementioned data to estimate origin-destination travel demand on the road network. Inverse Reinforcement Learning (IRL) models were explored as novel way to derive route choice information from the observed vehicle trajectories, which can be used to improve the state-of-art origin-destination demand estimation models. Possible extensions of the IRL models are also discussed in order to deal with non-representative vehicle trajectories.

2:00pm - 3:00pm

Wednesday, 12 February 2020


Room L502
Hawken Engineering Building (50)