Human motion oriented control method for humanoid robot

Abstract

The human can move flexibly in spite of the redundant degree of freedom. That is because the human previously acquires "knack" or "skill" such as timing in moving joints. Since the structure of a humanoid robot is similar to that of human, it is effective to use the "knack" or "skill" for a humanoid robot. The purpose of this research is to develop the control method to make the humanoid robot execute a human motion such as "knack" or "skill". As data for acquiring "knack" or "skill", we acquire angle patterns, myoelectricity patterns, time-varying images. The myoelectricity patterns is converted to a torque patterns. We analyze data of human motion in view of relations between movement of a joint angle and a direction of torque. In this research, we analyze a walking motion. We divide the walking motion into the part of control of COG by supporting legs and the part of the swing motion which human doesn't give torque. We control COG with Jacobian matrix and ZMP. In this research, we regard the motion given no torque when human swings a leg as "knack" or "skill". We apply this "knack" or "skill" to humanoid model in simulator OpenHRP and realize the motion of humanoid robot by using human's "knack" or "skill".