Integration of Robust Control with Reinforcement Learning for Safe Autonomous Vehicle Motion

This paper presents a control design framework for the integration of robust control and reinforcement learning-based (RL) control agent. The proposed integration method is applied for motion control of autonomous road vehicles, providing safe motion. In the integrated motion control, longitudinal and lateral dynamics are incorporated. The high-performance motion of the vehicle, e.g., high-velocity motion, path following, and reduction of lateral acceleration, through the RL-based control agent is achieved. The training through Proximal Policy Optimization during episodes is performed. Safe motion with guaranteed performances, i.e., keeping limits on lateral error, through the robust control and the supervisor is achieved. The robust control is designed through the H∞ method, and in the supervisor, a constrained quadratic programming task is performed. As a result, lateral and longitudinal control inputs of the vehicle are calculated by the integrated control system. The effectiveness of the proposed control method using simulation scenarios and test scenarios on a small-scaled test vehicle is illustrated.