Effects of Prior Knowledge on Multi-Agent Reinforcement Leaning System to Find Courses of Ships
Takeshi Kamio, Kunihiko Mitsubori, Takahiro Tanaka, Hisato Fujisaka, Kazuhisa Haeiwa · International Conference on Intelligent Information Processing · 2010
Ship transportation is important in the countries with sea and wide rivers. Multi-ship course problem has been treated in the engineering related to ships. But, the optimality of courses and the interaction between maneuvering actions have not been sufficiently discussed yet. Since there are the special conditions in ship maneuvering, we regard multi-agent reinforcement learning system (MARLS) as a useful tool to brisk up these discussions. In this paper, we propose a new MARLS to find courses of ships and investigate the effects of prior knowledge on our MARLS. From numerical simulations, we show that our MARLS can keep navigation rules and improve the learning efficiency.