Self learning of gravity compensation by LOCH humanoid robot

A humanoid robot is a complex machine with many degrees of freedom. And, the control at the joint level is a crucial step for a humanoid robot to achieve fast and accurate movements. In this paper, we address the issue of gravity compensation, and propose a learning approach which is inspired by a human-like scheme of compensating gravity through learning. First of all, we will describe the importance of gravity compensation. Then, for the purpose of comparison, we outline the theoretical way of computing the torques which compensate the gravity acting on a limbpsilas links and payload. Subsequently, we present a human-like learning scheme, which accurately determines the necessary torques for the compensation of gravity acting at the various joints when a humanoid robot is in any posture of interest. Finally, real experiments with a humanoid robot are presented and discussed.