Reinforcement Learning-based Traffic Control with Multiple Interacting Connected and Autonomous Vehicles

This paper develops a centralized reinforcement learning (RL)–based controller for Connected and Automated Vehicles (CAVs) operating in mixed autonomy settings designed to lower overall traffic fuel consumption. The traffic dynamics are formulated as a coupled PDE–ODE system, with the PDE governing macroscopic density evolution of the traffic and the ODEs characterizing the motion of CAVs that act as moving actuators. We integrate finite volume numerical schemes with the RL controller as the control input. Numerical experiments demonstrate that the proposed RL controller achieves a lower fuel consumption rate compared to the benchmark global optimization and Model Predictive Control (MPC) approaches. The results highlight the potential of data-driven control strategies for improving energy efficiency and sustainability in future mixed traffic systems.