Motion Capture Based Model Identification of the Humanoid Robot REEM-C Using Static Poses

Abstract

In this paper, we describe an approach for the model identification of the humanoid robot REEM-C. In contrast to previous work, we do not attempt to determine all dynamic parameters simultaneously. It is not clear whether such approaches can lead to redundancies in the optimization problem. We deliberately restrict ourselves to a very precise determination of the center-of-mass (COM) and the mass of the individual rigid bodies. As a result, we do not use Persistent Exciting (PE) trajectories and perform the identification based on motion capture and force plate measurements of 172 static poses. This results in more accurate experimental data and allows a more precise update of static parameters by means of an optimization problem. The inertial parameters are not updated and have to be adjusted using classical approaches, but based on the already improved static parameters. We report the performance of optimization by comparing the distance of the ground-projected-center-of-mass (GCOM) against the measured GCOM from the model information of the original and optimized model for each static pose. The improvement of the optimized model is furthermore reflected by means of a recorded dynamic squat motion and by analyzing the residual torques and forces acting at the floating base of the robot.