Imitation of human motion on a humanoid robot using non-linear optimization

In this paper, we present a system for the imitation of human motion on a humanoid robot, which is capable of incorporating both vision-based markerless and marker-based human motion capture techniques. Based on the so-called Master Motor Map, an interface for transferring motor knowledge between embodiments with different kinematics structure, the system is able to map human movement to a human-like movement on the humanoid while preserving the goal-directed characteristics of the movement. To attain an exact and goal-directed imitation of an observed movement, we introduce a reproduction module using non-linear optimization to maximize the similarity between the demonstrated human movement and the imitation by the robot. Experimental result using markerless and marker-based human motion capture data are given.