Lane-keeping Control of Autonomous Vehicles Using Reinforcement Learning and Predictive Safety Filter ^*

The urgent need for both safety and high-performance in the motion planning and decision-making system of the autonomous vehicle presents a significant challenge for learning-based technologies. In this paper, we introduces a lane-keeping control structure that combines reinforcement learning and a predictive safety filter to ensure the vehicle’s safety motion. A vehicle agent under the racetrack environment is trained by the soft actor critic algorithm, to achieve high-performance continuous lane-keeping motion. Additionally, we establish the safety task for lane-keeping, which results in a learning-based safety filter, to avoid violation of driving on the outer lane. Through simulations, we demonstrate the effectiveness of our method in ensuring the safe and trustworthy control of autonomous vehicle lane-keeping.