A distributed, collective intelligence framework for collision-free navigation through busy intersections

Connected autonomous vehicles have the potential to cooperate at a higher level than human drivers, for instance to navigate busy intersections with relaxed traffic rules. This paper presents a novel system for autonomous traffic management through intersections based on distributed cooperative intelligence. In this framework each vehicle decides its own motion, rather than entrusting control of all vehicles to a central agent which then becomes a single point of failure. The motion plans are developed by reinforcement learning on simulations of intersection traffic. The developed system is shown to route vehicles to their destination in a safe and timely manner. In the near future, this system can be deployed to as a collision warning/collision avoidance aid to human drivers by selectively acting on the discrepancy between a human's behavior and the AI-determined safe behavior.