Computational understanding of learning-by-demonstration process to acquire fast whole body motions sequentially from the low speed motions

2017 Fiscal Year Final Research Report

• Project/Area Number: 26280098
• Research Category: Grant-in-Aid for Scientific Research (B)
• Allocation Type: Partial Multi-year Fund
• Section: 一般
• Research Field: Intelligent robotics
• Research Institution: Ritsumeikan University
• Principal Investigator: Hyon Sang-Ho 立命館大学, 理工学部, 准教授 (30344691)
• Co-Investigator(Kenkyū-buntansha): 松原 崇充 奈良先端科学技術大学院大学, 情報科学研究科, 准教授 (20508056) ; 大塚 光雄 立命館大学, スポーツ健康科学部, 助教 (20611312) ; 下ノ村 和弘 立命館大学, 理工学部, 准教授 (80397679) ; 有木 由香 立命館大学, 理工学部, 助教 (80553239)
• Project Period (FY): 2014-04-01 – 2018-03-31
• Keywords: ヒューマノイドロボット / 運動制御 / 運動学習

Outline of Final Research Achievements

In this project, in order to clarify the teaching method of whole body movement with high degree of difficulty, we aimed to propose a teaching algorithm by reinforcement learning and to verify it using real humanoid robots. During the research period, a new teaching algorithm based on stochastic optimum control theory called Kullback-Leibler control was devised. We have experimentally proved the effectiveness of the proposed method using a dual-arm manipulator developed in this project. We also proposed to use a shared latent space between the human demonstrator and the robot, to quickly acquire the optimal control policy in the low-dimensional space. The method was also validated in dynamic simulations on a biped humanoid robot.

Free Research Field

ロボティクス