Research on intelligent collision avoidance decision-making of unmanned ship in unknown environments

To solve the problem of intelligent collision avoidance by unmanned ships in unknown environments, a deep reinforcement learning obstacle avoidance decision-making (DRLOAD) algorithm is proposed. The problems encountered in unmanned ships' intelligent avoidance decisions are analyzed, and the design criteria for a proposed decision algorithm are put forward. Based on the Markov decision process, an intelligent collision avoidance model is established for unmanned ships. The optimal strategy for an intelligent decision system is determined through the value function which maximizes the return for the mapping of the in unmanned ship's state to behavior. A reward function is specifically designed for obstacle avoidance, approaching a target and safety. Finally, simulation experiments are carried out in multi-state obstacle environments, demonstrate the effectiveness of the proposed DRLOAD algorithm.