01334615

Component

This paper presents a novel approach to validate a traffic simulation network. The use of reinforcement learning approach is studied for modeling vehicles' gap acceptance decisions at a stop controlled intersection. The proposed formulation translates a simple gap acceptance decision into a reinforcement learning problem, assuming that drivers' ultimate objective in a traffic network is to optimize wait-time and safety. Using an off-the-shelf simulation tool, drivers are simulated without any notion of the outcome of their decisions. From multiple episodes of gap acceptance decisions, they learn from the outcome of their actions i.e., wait-time and safety. A real-world traffic circle simulation network developed in Paramics simulation software is used to conduct experimental analyses. Results show that using Q-learning, a reinforcement learning algorithm, drivers' gap acceptance behavior can easily be validated at a high level of accuracy without extensively relying on field data.

01329018

11-3394

English

16p

2011

Transportation Research Board 90th Annual Meeting

DVD

Figures (9) ; References (23) ; Tables (2)

Decision making; Driving simulators; Gap acceptance; Highway operations; Traffic data; Traffic safety; Traffic simulation; Validation; Waiting time

Real world traffic; Reinforcement learning; Stop controlled intersections

Highways; Operations and Traffic Management; Planning and Forecasting; I72: Traffic and Transport Planning

Transportation Research Board Annual Meeting 2011 Paper #11-3394

TRIS, TRB

Feb 17 2011 6:30PM