The Proceedings of JSME annual Conference on Robotics and Mechatronics (Robomec)
Online ISSN : 2424-3124
2014
Session ID : 1A1-J01
Conference information
1A1-J01 Study on Symmetry under Locomotion-pattern Acquisition of Quadrupedal Robot Using Reinforcement Learning(Evolution and Learning for Robotics)
Satoru MATSUMOTOTetsuro YABUTA
Author information
CONFERENCE PROCEEDINGS FREE ACCESS

Details
Abstract
In our laboratory, we have succeeded in acquiring forward actions to various biological robot systems using Reinforcement Learning. And we have discussed biotic wonder by using robots that is easy to research. Therefore, this research attempts to reproduce in a quadrupedal robot the phenomenon that quadrupeds change gait from a walk to a trot to a gallop in response to running speed. In the past it has become clear that the acquired locomotion-patters have optimum velocity in point of energetic efficiency. On this point, a robot emulate to quadrupeds. But, in terms of the whole reinforcement learning result, acquired locomotion-patterns of robot are different from quadruped's. So this paper focuses attention on the symmetry of action and studies locomotion-pattern acquisition of a quadrupedal robot.
Content from these authors
© 2014 The Japan Society of Mechanical Engineers
Previous article Next article
feedback
Top