Ship cooperative collision avoidance strategy based on multi-agent deep reinforcement learning

Ship collision avoidance is the primary issue in intelligent navigation. In multi-ship encounters, only by collaborating and jointly planning collision avoidance strategies, the collision risk can be effectively reduced. In order to make the ship intelligent collision avoidance strategy collaborative, safe and practical, a ship collaborative collision avoidance decision method based on multi-agent deep reinforcement learning is proposed. Firstly, the method of identifying ship encounter situations is studied and a multi-ship collision avoidance strategy that satisfies the "International regulations for preventing collisions at sea" is designed. Secondly, by analysing the cooperation mode of multi-ship agents, a multi-ship agent cooperative collision avoidance decision-making model is constructed. The model uses the attention inference method to extract the key data that is helpful for collision avoidance decisions. And a memory driven experience learning method is designed to effectively accumulate interactive experience. In addition, the noise network and multihead attention mechanism are introduced into the model to enhance decision-making and exploration capabilities of ship agents. Finally, on the experimental map and the real nautical chart, simulation experiments are carried out on the multi-ship encounter scenarios. The results show that in terms of collaboration and safety, compared with multiple comparison methods, competitive results are obtained and the practical requirements are met using the proposed method, which provides a new solution for improving theintelligent navigation of ships and ensuring navigation safety. © 2023 Northeast University. All rights reserved.