Can Multi-DoF Training Improve Robustness of Muscle Synergy Inspired Myocontrollers?

Research output: Chapter in Book/Report/Conference proceedingConference contributionScientificpeer-review

Researchers

Research units

  • Imperial College London

Abstract

Non-negative Matrix Factorization (NMF) has been effective in extracting commands from surface electromyography (EMG) for the control of upper-limb prostheses. This approach enables Simultaneous and Proportional Control (SPC) over multiple degrees-of-freedom (DoFs) in a minimally supervised way. Here, like with other myoelectric approaches, robustness remains essential for clinical adoption, with device donning/doffing being a known cause for performance degradation. Previous research has demonstrated that NMF-based myocontrollers, trained on just single-DoF activations, permit a certain degree of user adaptation to a range of disturbances. In this study, we compare this traditional NMF controller with its sparsity constrained variation that allows initialization using both single and combined-DoF activations (NMF-C). The evaluation was done on 12 able bodied participants through a set of online target-reaching tests. Subjects were fitted with an 8-channel bipolar EMG setup, which was shifted by 1cm in both transversal directions throughout the experiments without system retraining. In the baseline condition NMF performed somewhat better than NMFC, but it did suffer more following the electrode repositioning, making the two perform on par. With no significant difference present across the conditions, results suggest that there is no immediate advantage from the naïve inclusion of more comprehensive training sets to the classic synergy-inspired implementation of SPC.

Details

Original languageEnglish
Title of host publicationProceedings of the 16th IEEE International Conference on Rehabilitation Robotics, ICORR 2019
Publication statusPublished - 1 Jun 2019
MoE publication typeA4 Article in a conference publication
EventIEEE International Conference on Rehabilitation Robotics - Toronto, Canada
Duration: 24 Jun 201928 Jun 2019
Conference number: 16

Publication series

NameIEEE International Conference on Rehabilitation Robotics
PublisherIEEE
ISSN (Print)1945-7898
ISSN (Electronic)1945-791X

Conference

ConferenceIEEE International Conference on Rehabilitation Robotics
Abbreviated titleICORR
CountryCanada
CityToronto
Period24/06/201928/06/2019

    Research areas

  • Electrodes, Electromyography, Task analysis, Training, Training data, Signal to noise ratio, Muscles

Download statistics

No data available

ID: 36912643