Autonomous segmentation of motion primitive including muscular activation using variational Bayesian mixture of Gaussian

Abstract

Motion primitive is a functional unit of human motion and this can provide a robot control framework to understand human motion in a physical way. Autonomous primitive segmentation is to divide a sequence of motions into a set of proper segments. This paper proposes primitive segmentation algorithm using Bayesian mixture of Gaussian that can automatically choose the number of mixture components. Also to resolve the ambiguous nature of the muscular activations within the same joint positions, the segmentation is performed augmenting the surface electromyogram as muscular activations into the joint positions.