Traffic Flow Simulation Based on Driver's Model of Decision Making

Abstract

In this paper, we introduce a new model for driver's route selection. In the model, a driver can decide a route before driving but may change it dynamically while driving. To decide the route before driving, the driver can use a Q-value map that is a result of reinforcement learning from the road information. Experience of driving the route and information offered from outside (e.g. via car navigation system) can make a driver change the route while driving. From the result of evaluation experiments with a traffic data of real world, the traffic flow simulator with the model works in more than 90% accuracy. To show how the model can process information from outside, we carry simple experiment in which a navigation system tells a driver the fastest route in the course of driving. The simulator produce some reasonable result. © 2004, The Institute of Electrical Engineers of Japan. All rights reserved.