Towards trajectory-based latent space control for human-robot collaboration

Abstract

In manual assembly, human-robot collaboration (HRC) research activities aim to improve the manual handling of parts. For this purpose, latent space control has been investigated by several research groups. However, control problems have shown limited applicability for various frame-based latent space models. The objective of this work is to determine if positional control is principally possible for trajectory-based latent spaces. Therefore, we investigate how ‘reach’ trajectories that are constrained in the target position map to a latent space that has been derived from functional principal component analysis (fPCA). A multivariate normality test is conducted to analyze the target-constrained position in the latent space, followed by mapping between the latent and Cartesian spaces. Results show that for linear shifts in the target ‘reach’ position, Gaussian modes in the latent space move along a vector in the latent space. Additionally, Gaussian mixtures occur at the target point where different motion styles are chosen for reaching. Results suggest that positional control may prove feasible for the considered type of latent space.