Research on Unmanned Surface Vessel Intrusion Evasion Strategy Based on Deep Reinforcement Learning

This paper focuses on the scenario where unmanned vessels (UAVs) from the red team need to break through the interception of unmanned vessels from the blue team, utilizing deep reinforcement learning technology to accomplish the intrusion evasion task. The study proposes a strategy with Multi-Agent Deep Deterministic Policy Gradient (MADDPG) reinforcement learning framework at its core. The paper presents a method for designing the state space based on the relative positions between the blue team's unmanned vessel and each red team's unmanned vessel. The speed of red team's unmanned vessels is considered as the action space for each agent. Additionally, reward functions are designed based on the goals of the intrusion evasion task to incentivize the red team's unmanned vessels to adopt expected behaviors. This ensures that the red team's unmanned vessels, while avoiding interception by the blue team's unmanned vessels, can successfully carry out the task of striking the target.