Automatic collision avoidance of multiple ships based on deep Q-learning |
| |
| Affiliation: | 1. School of Navigation, Guangzhou Maritime University, No. 101, Hongshan San Road, Guangzhou, 510725, China;2. Kobe Ocean-Bottom Exploration Center, Kobe University, 5-1-1, Fukae-minamimachi, Higashinada-ku, Kobe, 658-0022, Japan;3. Graduate School of Maritime Sciences, Kobe University, 5-1-1, Fukae-minamimachi, Higashinada-ku, Kobe, 658-0022, Japan;4. National Research Institute of Fisheries Engineering, Fisheries Research and Education Agency, 7620-7, Hasaki, Kamisu-shi, Ibaraki, 314-0408, Japan;5. Marine Electrical Engineering College, Dalian Maritime University, No. 1, Linghai Road, Dalian, 116026, China;1. School of Navigation, Wuhan University of Technology (WUT), China;2. Hubei Key Laboratory of Inland Shipping Technology (WUT), China;1. Centre for Marine Technology and Ocean Engineering (CENTEC), Instituto Superior Técnico, Universidade de Lisboa, Portugal;2. National Engineering Research Center for Water Transport Safety (WTS Center), China;3. Intelligent Transport Systems Research Center, Wuhan University of Technology, China;4. Department of Marine Technology, Norwegian University of Science and Technology, Norway |
| |
| Abstract: | As the number of ships for marine transportation increases with the globalisation of the world economy, waterways are becoming more congested than before. This situation will raise the risk of collision of the ships; hence, an automatic collision avoidance system needs to be developed. In this paper, a novel approach based on deep reinforcement learning (DRL) is proposed for automatic collision avoidance of multiple ships particularly in restricted waters. A training method and algorithms for collision avoidance of ships, incorporating ship manoeuvrability, human experience and navigation rules, are presented in detail. The proposed approach is investigated not only by numerical simulations but also by model experiments using three self-propelled ships. Through the systematic numerical and experimental validation, it is demonstrated the developed approach based on the DRL has great possibility for realising automatic collision avoidance of ships in highly complicated navigational situations. |
| |
| Keywords: | Automatic collision avoidance Deep Q-learning Multiple ships Navigation rules Experimental validation |
| 本文献已被 ScienceDirect 等数据库收录！ |
|
