Cooperative Strategy Learning and Knowledge Evolution to Adapt to Dynamism in Unknown Cooperation and Environment in Multi-agent System

2022 Fiscal Year Final Research Report

• Project/Area Number: 21KK0206
• Research Category: Fund for the Promotion of Joint International Research (Fostering Joint International Research (A))
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 61030:Intelligent informatics-related
• Research Institution: Okayama University
• Principal Investigator: Uwano Fumito 岡山大学, 自然科学学域, 助教 (30880687)
• Project Period (FY): 2022
• Keywords: 進化計算 / 強化学習 / 未知環境 / 知識転移 / マルチエージェントシステム

Outline of Final Research Achievements

This study proposed a new method for learning by considering the similarity between sequential transition states and unknown situations in the environment with perceptual aliasing by constructing new knowledge from the parts of the knowledge. That method contributes the learning in an unknown situation such that the robots can adapt to the unknown situation using the past observed data, and adapt to the sequential aliased states using hierarchically structured knowledge for expressing the sequential patterns of the environment.

Free Research Field

機械学習

Academic Significance and Societal Importance of the Research Achievements

本成果は，エージェントの数や環境の変化によって未経験の状況に直面しても，自身の知識を最大限に利用してそれに追従し，目的を達成する上で最適な行動を学習可能であるという点において重要な成果である．それに加えて，学習結果を人間が解釈可能な知識として保存可能な点も，今後のロボット系研究において重要である．なお，本成果は当該分野のトップカンファレンスであるGECCO 2023において発表する予定である．