On-line learning of a feedback controller for quasi-passive-dynamic walking by a stochastic policy gradient method

Kentaro Hitomi, T. Shibata, Yoshitaka Nakamura, Shin Ishii

Abstract

A class of biped locomotion called passive dynamic walking (PDW) has been recognized to be efficient in energy consumption and a key to understand human walking. Although PDW is sensitive to the initial condition and disturbances, some studies of quasi-PDW, which introduces supplementary actuators, are reported to overcome the sensitivity. In this article, for realization of the quasi-PDW, an on-line learning scheme of a feedback controller based on a policy gradient reinforcement learning method is proposed. Computer simulations show that the parameter in a quasi-PDW controller is automatically tuned by our method utilizing the passivity of the robot dynamics. The obtained controller is robust against variations in the slope gradient to some extent.
BibTeX
@inproceedings{hitomi2005line,
  author = {Kentaro Hitomi and T. Shibata and Yoshitaka Nakamura and Shin Ishii},
  title = {{On-line learning of a feedback controller for quasi-passive-dynamic walking by a stochastic policy gradient method}},
  booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems},
  year = {2005},
  doi = {10.1109/iros.2005.1545258},
}

Loading related content...

Figure