Motor unit subset selection for scalable real-time interfacing
AbstractCurrent methods for motor unit (MU) based human-machine interfacing do not scale well with the expansion of output functionality. This is due to the high computational demands of the initial MU parameter extraction via decomposition of high-density surface electromyography recordings. We propose an alternative approach that relies on task-specific batch decomposition processes along with a MU subset selection step to address feature redundancy. Offline analyses were conducted using EMG and kinematics pertaining to 18 wrist/forearm motor tasks recorded from 11 able-bodied subjects. The mutual information-based minimal Redundancy Maximal Relevancy (mRMR) feature selection framework was tested and compared to Maximal Relevancy (MR) and two arbitrary selection methods. Subset MUs were then used for joint kinematics estimation corresponding to those 18 motor tasks by three different regressors. The mRMR selection scheme was found to retain MUs with the highest predictive power. When the portion of tracked MUs was reduced to 25%, regression accuracy decreased by only 3.5%.
How to Cite
D. Yeung, F. Negro, and I. Vujaklija, “Motor unit subset selection for scalable real-time interfacing”, MEC Symposium, Aug. 2022.
Myoelectric Control Algorithms