Learning bowing gesture with motion diversity by dynamic movement primitives

Abstract

Motion diversity is an important factor in resembling motion with human-likeness. In this paper, we propose a method by which a robot acquires gesturing skills with motion diversity. We measured the bowing gesture with a 3D capturing camera and used dynamic movement primitives (DMPs) in order to capture the distribution of kernel parameters. The results showed that the height of kernel parameters have a Gaussian distribution and that the trajectory regenerated with the calculated parameters had the desired motion diversity.