Cooperative Hunting of Unmanned Surface Vehicles via Multi-Agent Proximal Policy Optimization Algorithm

This paper proposes a cooperative hunting algorithm based on multi-agent reinforcement learning (MARL) to address the problem of cooperative hunting at sea involving an evasive target. First, a three-degree-of-freedom kinematic model for an Unmanned Surface Vehicle (USV) is developed, considering the effects of sea wind and waves. The conditions for successful cooperative hunting are also established. Secondly, a reward function is designed to account for multiple objectives, including target capture, collaborative hunting, and collision avoidance. Then, the Cooperative Hunting Multi-Agent Proximal Policy Optimization algorithm is introduced, utilizing a centralized training and distributed execution framework to train the optimal strategy. Finally, simulation results demonstrate the effectiveness of the proposed algorithm.