Whole-body human-to-humanoid motion transfer

Abstract

This paper explores movement transfer from humans to humanoid robots. Our goal is to produce human-like motions in humanoid robots by capturing human motions and transferring these motions to humanoid robots. One crucial problem in human-humanoid motion transfer is the discrepancy between human and humanoid robot dynamic models. This model discrepancy generally results in an unbalanced motion as well as unnatural movement when the human motions are utilized by humanoid robots. In this paper, balance adjustment and angular-momentum adjustment schemes are proposed and applied to minimize the effect of the model discrepancy. The proposed schemes yield balance humanoid motions with minimal-angular momentum at the center of mass (CoM). Experiments were performed to capture human motions, and computer simulations were conducted to illustrate the performance and consistency of the proposed approach.