Towards life-long learning of posture control for s-EMG prostheses

TL;DR

This paper explores the potential for developing a lifelong learning system for s-EMG based posture control in prostheses, aiming to maintain accurate classification over time with minimal retraining.

Contribution

It investigates the feasibility of creating a stable posture classification system that remains valid over time, reducing the need for frequent retraining in real-life prosthesis use.

Findings

Initial results suggest potential for long-term stable classification.

The amount of training data influences classification robustness.

Further research needed for real-world application.

Abstract

Surface electromyography (s-EMG) sensors are a promising way to control upper-limb prostheses. However a training session is necessary in order to set up the controller that will make s-EMG based movement possible. All data recorded during the training session are used by a machine learning algorithm to make a posture classification, that will allow the controller to distinguish each posture. The aim of this study is to investigate if it's possible to make a posture classification which can remain valid over time. The next step will be the study of how it varies depending on the amount of information submitted to it during the training session in view of real life everyday use of the upper-limb prosthesis.