2018 Fiscal Year Final Research Report
Generalization using properties of the real world and autonomous division of state-action space(Fostering Joint International Research)
Project/Area Number |
15KK0015
|
Research Category |
Fund for the Promotion of Joint International Research (Fostering Joint International Research)
|
Allocation Type | Multi-year Fund |
Research Field |
Intelligent informatics
|
Research Institution | Hosei University |
Principal Investigator |
ITO Kazuyuki 法政大学, 理工学部, 教授 (90346411)
|
Research Collaborator |
Rossiter Jonathan ブリストル大学, ブリストルロボット研究所, 教授
|
Project Period (FY) |
2016 – 2018
|
Keywords | 知能ロボット / ソフトロボット / 汎化 / 強化学習 / 多脚ロボット / インフラ点検 / 災害対応 |
Outline of Final Research Achievements |
In this study, we focused on properties of the real world, and we proposed a framework to realize adaptive behavior of a robot using properties of the real world. We developed a soft multi-legged robot based on the proposed framework, and we demonstrated that the developed robot could obtain locomotion patterns for given environments using reinforcement learning. We also confirmed that the obtained policy to realize the locomotion pattern is generalized and is applicable for other similar environments without additional learning.
|
Free Research Field |
知能ロボット
|
Academic Significance and Societal Importance of the Research Achievements |
従来の「汎化とは多数の情報や経験を統計的に処理して共通の性質を抜き出すこと」という枠組みに対する対案として,「実世界の性質を利用して様々な状況に適応可能な身体を構成すること」という新しい枠組みを提案した点において学術的意義は大きい. また,実際のロボットを用いてその有用性を検証しており,この技術は,今後,災害対応や,インフラ点検,農林業など,複雑な未知環境で働くロボットを実現する上での基礎となり,社会的貢献も大きいと考えられる.
|