2026-03-12T00:44:41Z https://fit.repo.nii.ac.jp/oai

oai:fit.repo.nii.ac.jp:00000424 2023-05-15T13:41:05Z 256:257:279

ナッシュQ学習における協調行動の生成 Emergence of Cooperative Action in Nash-Q Learning ナッシュQ ガクシュウ ニ オケル キョウチョウ コウドウ ノ セイセイ 北原, 頌士 谷川, 裕一 鶴岡, 久 007.1 Q learning Nash-Q learning grid world agent reward Q learning Nash-Q learning grid world agent reward application/pdf 論文(Article) The effect of Nash-Q learning algorithm has not yet been confirmed in multiple experiments. We adopted a 5×5 grid world in which two agents started from opposite lower corners and tried to reach their respective goal cell. Experiments showed performance differences between single agent Q-learning and Nash-Q learning. In the Nash-Q learning, both agents obtained similar accumulated re-wards; however, in the Q learning, each agent accumulated his reward differently. Findings of this re-search confirmed that when agents adopt Nash Q-learning to predict the other agent’s behavior, not only is the performance of the agents better than their performance when using single-agent Q-learning, but the emergence of the cooperative action can also be observed. 福岡工業大学 2007-09-30 jpn departmental bulletin paper VoR http://hdl.handle.net/11478/937 https://fit.repo.nii.ac.jp/records/424 02876620 AN10036974 福岡工業大学研究論集 40 1 15 20 https://fit.repo.nii.ac.jp/record/424/files/11478-937_p15鶴岡　久.pdf application/pdf 843.3 kB 2020-09-14