Pattern-preserving-based motion imitation for robots

This paper presents a new algorithm of encoding dynamic movements through pattern-preserving optimization by a physical robot. This research follows a recent robot programming approach called learning from demonstration in which the motion trajectory is learned from human demonstrations. The motivation of this work is to deal with major challenges in learning from demonstration such as embodiment mapping, generalization, adaptation, robustness to perturbations, stability, pattern-preserving, and parameter tuning. We propose a new method that can deal with those problems and present empirical results to support our insistence.