場の構築を目的としたマルチロボットの協調動作学習

抄録

This paper discusses a multi-agent reinforcement learning (MARL) in a multi-agent robot system (MARS) to get cooperative behaviors for field constructions such as an environment creation and an information field construction. To learn cooperative behaviors by a Q-learning in dynamic environments where the MARS operates, we propose a method to give appropriate rewards to agents by switching two learning expressions situationally. Simulation results show that all agents in MARS obtain cooperative behaviors for environment arrangement with performing mutual collision avoidance by the proposed method.