USV Target Interception Control With Reinforcement Learning and Motion Prediction Method

Abstract

In this paper, an unmanned surface vehicle (USV) target interception problem is studied with reinforcement learning (RL)-based method. In the proposed new structure, the proximal policy optimization (PPO) and proportional derivative (PD) are combined. First, the PD controller is used to predict the interception position. Then, the PPO algorithm is trained to control the USV, so that it can move quickly to the predicted position. By comparing with the traditional PPO algorithm, the simulation results verify that the proposed algorithm spends less time solving the problem of the USV interception of a moving target.