APF guided reinforcement learning for ship collision avoidance

With the accelerated melting of Arctic Sea ice, the research and application of Arctic shipping routes have also accelerated. Maritime transportation serves as the primary mode of transportation; however, transportation risks are substantial. The extreme weather conditions in the Arctic region elevate the likelihood of maritime accidents. To mitigate the risks associated with ships navigating through Arctic ice regions and to prevent collisions between ships and ice, this article proposes an APF-PPO algorithm for collision avoidance utilizing Artificial Potential Field (APF) guided reinforcement learning algorithms. The APF-PPO algorithm offers dual guidance in decision-making and reward. Specifically, during decision-making, APF guidance supplants random decision-making, and an APF reward function is devised to significantly reduce ineffective exploration in the early stages of reinforcement learning, thereby enhancing the convergence speed of the reward function. Based on the results of simulation experiments, the APF-PPO algorithm demonstrates the capability to empower intelligent ships to evade obstacles and navigate securely within intricate ice environments.