Summary
2011 International Symposium on Nonlinear Theory and Its Applications
2011
Session Number:A2L-D
Session:
Number:A2L-D2
A Multi-agent Reinforcement Learning Method for Acquiring the Sociality
Yasuo Nagayuki,
pp.136-139
Publication Date:2011/9/4
Online ISSN:2188-5079
DOI:10.34385/proc.45.A2L-D2
PDF download (134KB)
Summary:
In multi-agent environments, it is important that the agents have the "sociality". In this article, I propose a reinforcement learning framework, which is based on Q-learning, that the agent is able to learn the "sociality" in a multi-agent environment. In this framework, the agent learns to ignore the near goal, which is left for the other agent, and go toward the farther goal, if the agent judges that the decision is effective from the social viewpoint, but not from the agent's greedy viewpoint.