Latent Space Based Collaborative Motion Modeling from Motion Capture Data for Human Robot Collaboration

Abstract

Collaborative assembly operation is one of the current challenges regarding human robot collaboration (HRC). In this context, it is still unclear how robots and humans should behave in handling an object and anticipating mutual care. In many cases, the modeling of collaborative behaviors shows difficulties, which can be addressed by simplifying motion modeling techniques. In the current work, we propose a latent space approach that combines functional principal component analysis to derive low dimensional features with Gaussian mixture models to generate high-likelihood motion behavior estimates. This approach may increase agility in task planning and reduce programming difficulties in HRC.