Authors:
Tara Baldacchino
;
William Jacobs
;
Sean R. Anderson
;
Keith Worden
and
Jennifer Rowson
Affiliation:
University of Sheffield, United Kingdom
Keyword(s):
sEMG Signals, Finger Force Regression, Variational Bayes, Multivariate Mixture of Experts, Prosthetic Hand.
Related
Ontology
Subjects/Areas/Topics:
Artificial Intelligence
;
Biomedical Engineering
;
Biomedical Signal Processing
;
Data Manipulation
;
Health Engineering and Technology Applications
;
Human-Computer Interaction
;
Methodologies and Methods
;
Neurocomputing
;
Neurotechnology, Electronics and Informatics
;
Pattern Recognition
;
Physiological Computing Systems
;
Physiological Processes and Bio-Signal Modeling, Non-Linear Dynamics
;
Sensor Networks
;
Soft Computing
Abstract:
Improving the dexterity of active prostheses is a major research area amalgamating machine learning algorithms
and biosignals. A recent research niche has emerged from this- providing proportional control to a
prosthetic hand by modelling the force applied at the fingertips using surface electromyography (sEMG). The
publicly released NinaPro database contains sEMG recording for 6 degree-of-freedom force activations for
40 intact subjects. In this preliminary study the authors successfully perform multivariate force regression
using Bayesian mixture of experts (MoE). The accuracy of the model is compared to the benchmark set by
the authors of NinaPro; comparable performance is achieved, however in this work a lower dimensional feature
extraction representation obtains the best modelling accuracies, hence reducing training time. Inherent
to the Bayesian framework is the inclusion of uncertainty in the model structure, providing a natural step
in obtaining confidence bounds on the predi
ctions. The MoE model used in this paper provides a powerful
method for modelling force regression with application to actively controlling prosthetic and robotic arms for
rehabilitation purposes, resulting in highly refined movements.
(More)