| Abstract | There is a growing demand for autonomous navigation technologies capable of operating effectively in complex dynamic environments where ships interact with various ships and maritime obstacles. To avoid collisions and ensure safe navigation, the own ship (OS) must plan routes that consider the potential risks posed by surrounding target ships (TSs). Classical methods based on reciprocal velocity obstacles (RVO) offer the strength of providing a safe speed zone and enabling real-time motion control by selecting the optimal speed within this zone. However, due to their rule-based nature, these methods may result in overly conservative route planning or become vulnerable to inaccuracies in the positions of TSs. Furthermore, planned routes may oscillate or fail to avoid collisions in complex and dynamic environments with many variables. On the other hand, collision avoidance methods using deep reinforcement learning (DRL) show strength in handling complex environments due to their ability to learn and adapt. However, these methods may sometimes fail to find an optimal policy or converge to an unstable state, potentially compromising safety and efficiency. In this study, we proposed a method for collision avoidance that combines classical RVO-based methods with the DRL. We represented the states of the OS and TSs as RVO vectors and incorporated them into a DRL model. This DRL model, reflecting the characteristics of the RVO, can guide the OS to navigate towards its destination safely and efficiently while adhering to collision risk assessment metrics and the COLREGs (Convention on the International Regulations for Preventing Collisions at Sea). By applying the proposed method to various collision scenarios in virtual environments, the results showed that even in complex maritime environments with surrounding TSs, the OS can reliably avoid collisions and navigate toward its intended destination. |
|---|---|
| Publication Date | 2025-06-05 |
Seong-Won Choi, Myung-Il Roh, In-Chang Yeo, "A Method for Collision Avoidance of an Autonomous Ship in Dynamic Environments", Proceedings of ISOPE(International Society of Offshore and Polar Engineers) 2025, Goyang, Korea, p. 3660-3663, 2025.06.01-06