A Path Tracking Control Method for Unmanned Surface Vehicle Based on Reinforcement Learning and Sliding Mode

This paper presents an integrated guidance and control approach for path tracking of unmanned surface vehicle (USV). To overcome the limitations of conventional geometric guidance methods, such as singularities and poor adaptability, a deep reinforcement learning framework is developed to optimize both guidance and heading control. Specifically, the deep deterministic policy gradient algorithm is enhanced with an action differential limitation mechanism, which effectively reduces output oscillations and improves operational safety. For low-level control, an integral sliding mode controller is designed to accurately track the heading and speed commands generated by the guidance layer, while explicitly accounting for USV dynamics and ensuring system stability. Simulation results demonstrate that the proposed method achieves faster convergence and smaller overshoot compared to traditional line-of-sight guidance. The approach also yields smoother dynamic responses, contributing to improved tracking safety and performance.