Dynamics-Based Motion Synthesis by Simple Learning Control

Abstract

Point-to-point motion of a humanoid agent (or any mobile creature) is considered as a composition of basic movements related to the degrees of freedom. With time/energy performance criteria, such movements can be synthesized by using simple control functions. Natural looking animation can be achieved employing even very simplified dynamics models. Their parameters can be estimated or identified using motion capture data. In case no appropriate dynamic model is available, we propose an efficient procedure of direct motion editing. The captured motion can be re-used in order to animate the original or other characters in a variety of similar motion tasks. The proposed approach for motion synthesis is applied in the case of a six-link biped. Two examples are considered: one for editing captured walking motion and the other of direct motion synthesis for climbing stairs. The proposed methodology is very appropriate for interactive character animation.