Summary
International Symposium on Nonlinear Theory and Its Applications
2022
Session Number:C5L-E
Session:
Number:C5L-E-03
Identification of Avoidance Starting Points by Reinforcement Learning-Based Multi-Ship Course Search Method with Target Courses as Actions
Takeshi Kamio , Hiroki Kimura , Takahiro Tanaka , Kunihiko Mitsubori , Hisato Fujisaka,
pp.589-592
Publication Date:12/12/2022
Online ISSN:2188-5079
Summary:
Since navigation rules (NRs) only roughly define how to avoid collisions between two ships, the actual navigators must make decisions about the direction and timing for avoidance based on their experience. Therefore, the decisions of the unskilled navigators tend to be ambiguous. Against this background, we have discussed course efficiency and safety using a multi-agent reinforcement learning system (MARLS) to search ships' courses. However, we have not discussed avoidance timing. In this paper, we propose a method to identify avoidance starting points using our MARLS. Through numerical experiments, we have confirmed that our proposed MARLS can find efficient courses and converge the avoidance starting points corresponding to each avoidance to a small area.