Steering Autonomous Driving Agents Through Intersections in Virtual Urban Environments

This paper explores how autonomous driving agents drive through intersections in virtual urban environments. The knowledge about where a vehicle runs on an intersection is embedded in environment database. Control of how and when a vehicle runs through an intersection is provided by vehicle behaviors. Vehicle control for traversing an intersection is divided into cruising behavior, following behavior, and intersection behavior. The component behaviors are integrated together to steer an autonomous driving agent through intersections. Before a virtual vehicle enters an intersection, the driving agent should make a decision that the vehicle either goes forward to pass the intersection or stops before the intersection. The chosen action is then applied to vehicle control. Because the ambient traffic and the status of traffic control signals on the intersection are dynamic, the decision should also be made on each time step.