Explainable AI for ship collision avoidance: Decoding decision-making processes and behavioral intentions

Most ship collision accidents are attributed to human errors. Autonomous navigation technology is heralded as a potential solution to mitigate human error-related collisions. Recent advancements have enabled the application of deep reinforcement learning (DRL) in developing autonomous navigation artificial intelligence (AI). However, the decision-making process of AI is not transparent, and its potential for misjudgment could lead to severe accidents. Consequently, the explainability of DRL-based AI emerges as a critical hurdle in deploying autonomous collision avoidance systems. This study developed an explainable AI for ship collision avoidance. Initially, a critic network composed of sub-task critic networks was proposed to individually evaluate each sub-task to clarify the AI decision-making processes in collision avoidance. Additionally, an attempt was made to discern behavioral intentions through a Q-value analysis and an Attention mechanism. The former focused on interpreting intentions by examining the increment of the Q-value resulting from AI actions, while the latter incorporated the significance of other ships in the decision-making process for collision avoidance into the learning objective. AI's behavioral intentions in collision avoidance were visualized by combining the perceived increment of Q-value with the degree of attention to other ships. The proposed method was evaluated through a numerical experiment. The developed AI was confirmed to be able to safely avoid collisions under various congestion levels, and the decision-making process and behavioral intention of AI for collision avoidance were rendered comprehensible to humans. It is comprehensible to seafarers onboard and could contribute to the future practical implementation of autonomous navigation AI systems.