Interaction Mesh Based Motion Adaptation for Biped Humanoid Robots

Adapting human motion data for humanoid robots can be an efficient way to let them conduct specific tasks or perform visually intriguing movements. However adapting movements which involve close interactions between body parts / objects has been a difficult problem if we try to make the motions sufficiently similar to the original ones without causing undesired collisions and fall-overs. To solve this problem, we propose a novel motion adaptation system which can adapt full-body motions of a human to those by a biped humanoid robot. The system integrates the joint coordinate model and a dynamic balancing method with the interaction mesh based motion adaptation method, which is originally designed for character motions represented by marker positions in Cartesian coordinate. We present that the system can adapt human motion sequences that include close interactions between the arms for the biped humanoid robot HRP-4C, and the adapted motion can be stably performed in dynamics simulation.