Reinforcement Learning-Based Robust Control for Path Tracking of Automated Vehicles

This paper proposes a novel method to combine a reinforcement learning-based control agent and robust controller to provide a high-performance robust control solution. The proposed integration method is applied for motion control of autonomous road vehicles, providing safe motion. In the integration, motion control on longitudinal and lateral dynamics is performed. Depending on the chosen reward function different driving characteristics are achieved e.g., minimal lap time, accurate path following, or enhanced passenger comfort. The training of the neural network is performed using Proximal Policy Optimization. The robust controller is designed through \(\mathcal {H}_\infty \) method, and the two controllers are combined using a supervisor, which performs a constrained quadratic programming task. 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 small-scaled test vehicles is illustrated.