Action myoelectric control for advanced hand prostheses via multi-label classification
We propose action control, a novel approach for myoelectric independent digit control based on multi-label classification. At each time step, the decoder classifies movement for each controllable degree-of-freedom (DOF) into one of three categories: open, close or stall (i.e., no movement). The user employs continuous feedback information to estimate and minimise the mismatch between target and current digit positions. We implemented the proposed action controller and evaluated its real-time performance with 3 transradial amputee—two bilateral, one unilateral—, whilst they controlled a six-dimensional computer interface with surface electromyography (EMG) signals. We benchmarked the performance of the algorithm against the state-of-the-art in myoelectric digit control, that is, position control using multi-output regression. We found that action control consistently and substantially outperformed position control. Furthermore, all participants rated action higher than position control in a series of questions in a post-experimental survey and expressed and overall preference for the former. The proposed algorithm warrants further investigation in the future by transferring the control space from a computer display onto a real prosthesis and evaluating performance during activities of daily living.