Synthetic Robotics Approaches for Understanding the Formation and Deformation of Cooperation between Agents Based on Predictive Coding

2022 Fiscal Year Final Research Report

• Project/Area Number: 19K20364
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 61040:Soft computing-related
• Research Institution: Keio University (2020-2022); National Institute of Informatics (2019)
• Principal Investigator: Murata Shingo 慶應義塾大学, 理工学部(矢上), 講師 (80778168)
• Project Period (FY): 2019-04-01 – 2023-03-31
• Keywords: 予測符号化 / 予測誤差最小化 / 認知ロボティクス / ニューラルネットワーク / インタラクション

Outline of Final Research Achievements

The aim of this research project is to understand the cognitive information processing mechanisms that support cooperation with others through synthetic robotics approaches integrating cognitive neuroscience, machine learning, and robotics. In particular, the focus is on the dynamic processes of the formation and deformation of cooperation, which are influenced by external factors such as other agents and the environment, as well as internal factors such as one's own intentions and goals. We have developed computational models based on predictive coding and conducted a set of robot experiments. Specifically, we have proposed a gradient-based optimization method and a more accelerated amortized inference method. We validated deep generative models equipped with these methods and applied them to collaborative robots to evaluate their performance in terms of both fundamental and practical aspects.

Free Research Field

認知ロボティクス

Academic Significance and Societal Importance of the Research Achievements

本研究成果は，他者や環境といった外的要因と自己の意図やゴールといった内的要因によって生じる協調の形成とその崩壊が，予測符号化という単一の仕組みによって説明可能であることを示した．具体的には，過去に生じた予測誤差を最小化することで外的要因を知覚し，未来に生じると想定される予測誤差を最小化することで内的要因を満たす行動生成が可能であることが確認された．本研究で得られた成果は他者との協調のみならず，共同注意や心の理論といった社会性認知に関する問題への貢献も期待される．また，ロボットを含む機械による他者（人間や他の機械）との円滑な協調の実現へと繋がる工学的応用可能性も期待される．