A Promoting Method of Role Differentiation using a Learning Rate that has a Periodically Negative Value in Multi-agent Reinforcement Learning
Masato Nagayoshi, Simon Elderton, Hisashi Tamaki · Journal of Robotics Networking and Artificial Life · 2020
There have been many studies on Multi-Agent Reinforcement Learning (MARL) in which each autonomous agent obtains its own control rule by Reinforcement Learning (RL). Here, we hypothesize that different agents having individuality is more effective than uniform agents in terms of role differentiation in MARL. In this paper, we propose a promoting method of role differentiation using a wave-form changing parameter in MARL. Then we confirm the effectiveness of role differentiation by the learning rate that has a periodically negative value through computational experiments.