Learning the Latent Space of Robot Dynamics for Cutting Interaction Inference

TL;DR

This paper explores using variational autoencoders to learn a low-dimensional latent space of robot cutting dynamics, enabling inference of cutting states, material, and geometry with competitive prediction accuracy.

Contribution

It introduces two VAE models for capturing complex robot interaction dynamics and demonstrates their effectiveness in inference tasks compared to recurrent neural networks.

Findings

Latent space effectively captures cutting interaction properties.

VAE models outperform some RNN-based approaches in prediction.

Latent representations are expressive for robotic environment interaction inference.

Abstract

Utilization of latent space to capture a lower-dimensional representation of a complex dynamics model is explored in this work. The targeted application is of a robotic manipulator executing a complex environment interaction task, in particular, cutting a wooden object. We train two flavours of Variational Autoencoders---standard and Vector-Quantised---to learn the latent space which is then used to infer certain properties of the cutting operation, such as whether the robot is cutting or not, as well as, material and geometry of the object being cut. The two VAE models are evaluated with reconstruction, prediction and a combined reconstruction/prediction decoders. The results demonstrate the expressiveness of the latent space for robotic interaction inference and the competitive prediction performance against recurrent neural networks.