A Cooperative DRL Approach for Autonomous Traffic Prioritization in Mixed Vehicles Scenarios

The number of connected and automated vehicles in urban areas will gradually increase in the near future. As a consequence, mixed traffic made of both regular human- driven and connected and automated vehicles will likely be a typical scenario over the next few years. Connected and automated vehicles can benefit the whole traffic experience by preventing collisions and optimizing traffic waves. In this context, unsignalized intersection management is one of the most critical challenge. However, it has received less attention in recent literature. In this paper, we introduce a novel approach to optimize traffic flow at unsignalized intersections by using cooperative deep reinforcement learning. In the proposed multi- agent strategy, each connected automated vehicle (CAV) acts as an agent and has partial observability of the current state made of other CAV s, regular vehicles and priority connected vehicles (CPV), such as ambulances and police cars. The ultimate goal of each agent is to give right-of-way to priority vehicles, prevent collisions and optimize intersection crossing time. Experimental results of the simulations carried out in SUMO on randomly generated traffic validate the presented approach and show the performance and the safety of the final trained policy.