Exploring user compliance in the training of regression-based myoelectric control
DOI:
https://doi.org/10.57922/mec.2507Abstract
Regression-based myoelectric control is promising because it could enable simultaneous independent control over multiple degrees of freedom. However, limitations in the robustness of online control and the added complexity of acquiring labelled training data have hindered its adoption. The scattered mix of prompting methods, visualization styles, and speeds used in the literature to collect labelled regression training data has obfuscated how these different prompting styles may affect the performance of regression-based myoelectric control. This work thus begins to investigate the potential effects that different visualizations may have on user behaviours and the quality of acquired training data for myoelectric control. Two distinct behaviours, referred to as all or nothing and anticipation, emerged when comparing the training data of three different prompting styles. Subsequently, 6 subjects were coached to emulate each of these behaviours during training and then completed a 10-trial Fitts' Law to assess the online usability of two different support vector regression (SVR) models. Results show that both user behaviours and the choice of regression model can have profound impacts on the usability of regression-based myoelectric control. Notably, real-time performance was severely degraded by the anticipation behaviour when using a linear kernel SVR, resulting in a 70% reduction in completion rate. These preliminary results motivate future work into how best to prompt users when training for supervised regression-based myoelectric control.
Downloads
Published
2024-08-15
How to Cite
[1]
C. Morrell, E. Campbell, and E. Scheme, “Exploring user compliance in the training of regression-based myoelectric control”, MEC Symposium, Aug. 2024.
Conference Proceedings Volume
Section
User Experience
