Training an EMG-Based Machine Learning Model to Classify Hand Gestures in a Spatial Virtual Reality Environment
Authors
Olsen, Tóki Lava ; Johansen, Charlotte ; Sørensen, Kristian Hedegaard
Term
4. term
Education
Publication year
2025
Submitted on
2025-05-27
Pages
14
Abstract
This study explored the use of an electromyography (EMG) based machine learning model to classify hand gestures in a spatial virtual reality (VR) environment. A total of 18 participants participated in the evaluation. The EMG signals were recorded from the forearm of each participant, and they were used to train intra-subject classification models to predict the movement of a hand prosthetic. The results showed a significant gap between the offline and online performance of the trained model, with macro F1 scores averaging 0.86, while a notable average performance drop was observed during the online tests, where the macro F1 score fell to 0.52. Individual participants achieved more satisfactory scores, suggesting the presence of individual differences that may be influenced by various contributing factors. The results show promising steps towards training and testing a machine learning algorithm to control a hand model with EMG signals in VR. With further development, this approach has the potential to support amputees in training a prosthetic hand within a spatial VR environment prior to receiving the physical device
Keywords
Documents