Research on Lateral Adaptive Control Method of Unmanned Vehicle Based on Reinforcement Learning

In order to solve the problems that the fixed parameters of the traditional path tracking control algorithm cannot meet all the path tracking control requirements, and the parameter tuning is highly dependent on experience and will consume a lot of time and energy, this paper proposes a lateral adaptive path tracking control method for unmanned vehicles based on reinforcement learning. A two-layer controller is designed for path tracking control. The lower controller combines multi-point preview, pure-pursuit control and PID to calculate the expected front wheel angle; the upper controller uses the soft-Actor Critic (SAC) algorithm to adaptively tune the parameters in the lower controller, and designs the state space, reward function and action space for the path tracking problem. The combination of the two not only ensures the safety of the reinforcement learning model in the path tracking control process, but also effectively improves the convergence speed of the reinforcement learning model training. The model is built in Prescan for training and simulation testing, and a real vehicle verification is carried out in an open-pit mine area to verify the effectiveness of the proposed method.