Adaption of performed ballistic motion

Abstract

This thesis presents a method for adapting performed ballistic motions of a full human figure with many degrees of freedom by using an optimal trajectory formulation and the dynamics derived from first principles. Computation of the joints, torques, and reaction forces allows the application of a number of optimization criteria that result in creation of natural looking final motion. Alternatively, a reduced-order dynamics constraints can improve solution time by an order of magnitude and still retain the natural quality of the resulting motion in most adaptation scenarios. The adaptation method generates over twenty different adaptations from the original performances of a human jump and run. Although these results demonstrate the robustness of this method for a full human figure motion adaptation, an automated skeleton simplification is also presented. Applying common model reduction techniques, such as principal and independent component analysis, to the original motion data yields a low-dimensional character representation of a given motion activity. While the reduced character configuration converges faster for some optimization formulations, the high-dimensional character optimization always produces more natural looking motions.