Continuous Gesture Recognition from sEMG Sensor Data with Recurrent Neural Networks and Adversarial Domain Adaptation

TL;DR

This paper demonstrates that recurrent neural networks, especially SRUs, combined with domain adaptation techniques, significantly improve continuous gesture recognition accuracy from sEMG data for both mobile and non-mobile wrists, enhancing prosthetic control.

Contribution

It provides the first empirical results on gesture recognition with mobile wrists and shows that SRUs outperform regular RNNs, with domain adaptation improving cross-subject transfer.

Findings

SRUs outperform regular RNNs in gesture recognition accuracy.

Domain adaptation enhances transferability of models between subjects.

Empirical results include both mobile and non-mobile wrist scenarios.

Abstract

Movement control of artificial limbs has made big advances in recent years. New sensor and control technology enhanced the functionality and usefulness of artificial limbs to the point that complex movements, such as grasping, can be performed to a limited extent. To date, the most successful results were achieved by applying recurrent neural networks (RNNs). However, in the domain of artificial hands, experiments so far were limited to non-mobile wrists, which significantly reduces the functionality of such prostheses. In this paper, for the first time, we present empirical results on gesture recognition with both mobile and non-mobile wrists. Furthermore, we demonstrate that recurrent neural networks with simple recurrent units (SRU) outperform regular RNNs in both cases in terms of gesture recognition accuracy, on data acquired by an arm band sensing electromagnetic signals from arm muscles (via surface electromyography or sEMG). Finally, we show that adding domain adaptation techniques to continuous gesture recognition with RNN improves the transfer ability between subjects, where a limb controller trained on data from one person is used for another person.