Application of Deep Reinforcement Learning for Traffic Control of Road Intersection with Autonomous Vehicles

The control of road intersection in the presence of priority vehicles is central in terms of the optimal management performance of the autonomous vehicle (AV) scenarios. In this chapter, two different deep reinforcement learning (DRL)-based approaches are presented for the effective management of signalized and unsignalized intersections. Firstly, a study applying DRL to traffic light control at a road intersection is addressed, also considering the presence of three classes of priority vehicles such as AVs, ambulances, and police. A case study of a road intersection in the city of Bari is presented. Secondly, a novel multiagent cooperative DRL approach is introduced for autonomous intersection management at unsignalized intersections under prioritized mixed traffic scenarios. In the proposed method, three different vehicle classes are considered: connected AVs (CAVs); connected priority vehicles (CPVs), such as ambulances or police cars, that are connected but driven by humans; and, lastly, regular vehicles (RVs) that are neither connected or automated. The chapter focuses on a high-level dynamics of traffic management, not considering low-level issues like communication and data transfer.