Feature Disentanglement of Robot Trajectories

TL;DR

This paper evaluates disentangled representation learning methods, specifically three VAEs, on robot trajectories, demonstrating that decorrelation-based models outperform others in disentangling and trajectory quality, advancing unsupervised learning in robotics.

Contribution

It introduces and evaluates a new $eta$-Decorr VAE and compares three disentangling VAEs on robot trajectory data, highlighting the effectiveness of decorrelation-based approaches.

Findings

Decorrelation-based VAEs outperform others in disentangling metrics.

Decorrelation VAEs produce higher quality trajectories.

Results support increased use of unsupervised learning in robot control.

Abstract

Modeling trajectories generated by robot joints is complex and required for high level activities like trajectory generation, clustering, and classification. Disentagled representation learning promises advances in unsupervised learning, but they have not been evaluated in robot-generated trajectories. In this paper we evaluate three disentangling VAEs ($\beta$-VAE, Decorr VAE, and a new $\beta$-Decorr VAE) on a dataset of 1M robot trajectories generated from a 3 DoF robot arm. We find that the decorrelation-based formulations perform the best in terms of disentangling metrics, trajectory quality, and correlation with ground truth latent features. We expect that these results increase the use of unsupervised learning in robot control.