1A1-M09 強化学習を用いたステッピングによる転倒回避動作の学習(人間機械協調)

概要

論文の詳細を見る
This paper presents a method to learn stepping motions for fall avoidance with reinforcement learning. In order to overcome the curse of dimensionality associated with the large number of degrees of freedom with a humanoid robot, we consider learning on the reduced dimension state space based on a simplified inverted pendulum model. A feedback controller is introduced in the ankle joints so that the center of mass is forced to follow the dynamics of the inverted pendulum model. The proposed method is applied to a humanoid robot in numerical simulations, and simulation results demonstrate the feasibility of the proposed method to acquire appropriate stepping motions to avoid falling over caused by external perturbations.
一般社団法人日本機械学会の論文
2007-05-11