Road traffic simulation is more and more used in a wide range of applications: impact analysis of new road infrastructure, driver support systems design and assessment, driving learning, training or retraining, etc. In this context, having a validated traffic simulation for urban situation is a real need. Simulating a urban network is a hard problem due to the important number of junctions with merging and/or inserting flow. Even if the driving task is competitive, simulated drivers have to coordinate themselves as in real life. First works allowed to define a multiagent coordination mechanism respecting constraints imposed by multi-agent simulation philosophy : distributed, non normative and generic. To avoid deadlocks at junctions, and as drivers do in actual situations, simulated drivers have to anticipate. Not only is anticipation realistic since it is a fundamental component of driving task, but anticipation can also be used to reinforce informations available for the multi-agent coordination mechanism. (Author/publisher)
Abstract