Do-Hyun Chun, Myung-Il Roh, Hye-Won Lee, "Deep Reinforcement Learning-Based Ship Collision Avoidance Considering Collision Risk", Proceedings of TEAM(Asian-Pacific Technical Exchange and Advisory Meeting on Marine Structures) 2022, Istanbul, Turkey, pp. 268, 2021.12.06-08
International Conference
2021.12.17 14:33
Do-Hyun Chun, Myung-Il Roh, Hye-Won Lee, "Deep Reinforcement Learning-Based Ship Collision Avoidance Considering Collision Risk", Proceedings of TEAM 2022, Istanbul, Turkey, pp. 268, 2021.12.06-07
조회 수 1221
첨부 '1' |
---|
Abstract | Recently, maritime accidents caused by human factors have increased according to the increase in the number of ships for sea transportation. Therefore, the autonomous navigation systems that find appropriate avoidance routes in a complex marine environment with various obstacles is attracting attention. In this study, we proposed a collision avoidance method using deep reinforcement learning (DRL) that generates an appropriate control action. DRL-based collision avoidance method derives the required rudder angle of own ship with given state of the own ship and target ship such as position, speed, and heading. To achieve the appropriate collision avoidance, it is necessary to assess the collision risk of the target ship accurately. Therefore, the probabilistic collision risk assessment method was proposed to predict the collision risk of the target ship with the probability distribution of the data. The probability distribution was calculated through the multivariate normal distribution of the four-dimensional variables of the position, speed, and heading angle of the target ship. The collision risk of the target ship was calculated through the probability distribution for each variable and the CPA (Closest Point of Approach)-based collision risk assessment method. To verify the proposed method, we applied the DRL-based collision avoidance method and the collision risk assessment method to various scenarios. The proposed method reliably avoided collisions through flexible paths for complex situations. |
---|---|
Publication Date | 2021-12-07 |