Whole-body motion retargeting using constrained smoothing and functional principle component analysis

Abstract

This paper presents a novel retargeting framework for humanoid robots that allows flexible motion representation and motion synthesis with smoothing with constraints and functional principle component analysis (Functional PCA). Constrained smoothing consists in computing base functions through optimizations with constraints including mechanical limits and stability conditions. By applying Functional PCA that is a statistic method describing given motions with principal components of functions, a variety of different motions can be expressed with a small number of parameters. We apply the proposed framework to whole-body “squat” motions to reveal that those motions can practically be classified with two components. The effectiveness of the proposed method is verified by dynamic simulations and experiments with the humanoid robot HRP-4.