Luman Zhao, Myung-Il Roh, Hye-Won Lee, Do-Hyun Chun, Sung-Jun Lee, "A Collision Avoidance Method of Multi-ships Based on Deep Reinforcement Learning Considering COLREGs", Proceedings of ICCAS(International Conference on Computer Applications in Shipbuilding) 2019, Rotterdam, Netherlands, pp. 85-88, 2019.09.24-26
International Conference
2019.08.06 15:07
Luman Zhao, Myung-Il Roh, Hye-Won Lee, Do-Hyun Chun, Sung-Jun Lee, "A Collision Avoidance Method of Multi-ships Based on Deep Reinforcement Learning Considering COLREGs," Proceedings of ICCAS 2019, Rotterdam, Netherlands, pp. 85-88, 2019.09.24-26
조회 수 1930
| 첨부 '1' |
|---|
| Abstract | Developing a high-level autonomous collision avoidance system for ships which can operate in an unstructured and unpredictable environment is a challenging task. Especially in the congested sea areas, each ship should continuously make decisions to avoid collisions with many other ships in the busy and complex waterway. Furthermore, recent reports indicate that a large number of collision accidents at sea are caused by or related to human decision failures with lack of situational awareness and failure to comply with International Regulations for Preventing Collisions at Sea (COLREGs). In this study, we propose a robust and efficient method to collision avoidance problems of multi-ships based on the deep reinforcement learning (DRL) algorithm. The proposed method directly maps the states of encountered ships to an ownship’s steering commands in terms of the rudder angle using a deep neural network (DNN). This DNN is trained over multi-ships on rich encountering situations using the policy gradient based DRL algorithm. To handle multiple encountered ships, we classify them into four regions based on COLREGs, and only consider the nearest ship in each region. We validate the proposed method in a variety of simulated scenarios thorough performance evaluations. The result shows that the proposed method can find time efficient, collision-free paths for multi-ships. Also, it shows that the proposed method has excellent adaptability to unknown complex environments. |
|---|---|
| Publication Date | 2019-09-24 |
