Mahalanobis-Taguchi-Fukuda Approach to Learning Human Motion Control

Abstract

Human motion trajectories vary widely from time to time to adapt to the situation. This is the case of one person. But as our body builds are different, motion trajectories vary in a wide variety of ways from person to person. Thus, human motion control is none other than the problem of tacit and nonverbal world. Usual explicit and verbal rational approaches are not effective. We need to find the best way to move by trial and error. But there is no performance indicator available. Usually we look at the supervising motion and try to follow it. But as our body builds are different, this does not work well. We must find our way by ourselves. Therefore, trial and error are essential and to improve our trial, we need a performance indicator. But Euclidian Space approach which is the basis of explicit and verbal rational approach does not work effectively due to its orthonormality and unit constraints. It is pointed out in this paper that if we introduce Mahalanobis Distance, which is a non-Euclidian Space approach free from orthogonality and unit constraints, we can establish a quantitative performance indicator and by coupling MD with a pattern approach, we can learn to control our motions by trial and error effectively.