Game of Marine Robots: USV Pursuit Evasion Game Using Online Reinforcement Learning

In this article, an online reinforcement learning (RL) algorithm is studied for the pursuit evasion game of Unmanned Surface Vehicles (USVs), both of which have learning abilities compared to the traditional apparent strategy. The pursuit evasion game between the USVs is described as differential game based on the relative motion equation to overcome the weakness of data-driven learning. The solution to this differential game is obtained by using online RL. The value function, the USV1 (pursuer) strategy, and the USV2 (evader) strategy are approximated by critic, actor 1, and actor 2 neural networks (NNs), respectively. The uniformly ultimately bound (UUB) of the system states and weight errors of NNs are researched based on Lyapunov theory. The performance of the proposed strategy is verified by the simulation results.