Cooperative Target Fencing of Uncertain Multi-USVs: A Reinforcement Learning-Based Approach

This article is dedicated to solving the cooperative collision-free target fencing control problem for a class of multiple unmanned surface vehicle (multi-USV) systems subject to external disturbances and uncertain dynamics. While existing control strategies often struggle to simultaneously address external disturbances, internal system uncertainties, and safety constraints in real-world applications, this study proposes a robust two-stage control strategy to effectively tackle these challenges. In the first stage, a baseline controller leveraging sliding mode control (SMC) is developed to effectively compensate for external disturbances and establish fundamental target fencing capabilities. Building on this foundation, the second stage introduces an advanced controller enhanced by soft actor–critic (SAC) reinforcement learning (RL) technique, which is specifically tailored to address internal system uncertainties and guarantee collision avoidance. Finally, numerical simulations are performed to validate the proposed approach, demonstrating its advantages over conventional RL-based methods in terms of efficiency.