Dynamic motion capture and edition using a stack of tasks

Abstract

This paper presents a complete methodology to quickly reshape a dynamic motion demonstrated by a human and to adapt the dynamics of the human to the dynamics of the robot. The method uses an inverse dynamics control scheme with a quadratic programming optimization solver. The motion data recorded using a motion capture system is introduced into the control scheme as a reference posture task to be followed by the joints trajectory respecting the dynamic limitations as well as the contact constraints. The motion is further modified using arbitrary tasks to let the robot imitate the original motion more closely or to make voluntary changes for aesthetic reasons. The results show the method applied to the humanoid robot HRP-2 imitating a human "pop dance".