Dynamic Joint Control for a Humanoid Walk

Worn-out robots are a topic that is often overlooked when developing approaches for walking with humanoid robots. When robots get older, their joints get more and more play, resulting in control commands being not executed as intended. In this paper, we present a novel approach for modifying the target joint angles of the leg joints of the humanoid robot NAO to bring its performance closer to a robot without considerable wear and tear. Based on the estimated center of mass and an estimated rotation error of the support foot, the joint velocities are adapted to handle stuck joints that result from the weight and movement of the robot. Moreover, the differences of the commanded joint angles and the measured ones are used to bring worn-out robots closer to the state of good ones by reducing such errors.