Evidence that a deep learning regression-based controller mitigates the limb position effect for an individual with transradial amputation

Abstract

Myoelectric upper limb prostheses provide wrist and hand movements to users yet remain somewhat unreliable and challenging to operate in high and cross-body limb positions. Hand and wrist movements are typically controlled sequen- tially and at a pre-set velocity. We have made significant inroads towards developing a novel controller that is reliable in multiple limb positions and offers fluid movements. Our recent work unveiled a promising deep learning regression- based myoelectric control solution. Herein we present results from our current study that tested device control using our solution versus a baseline (classification) alternative. A myoelectric prosthesis user with transradial amputation donned an experimental prosthesis and performed two functional tasks under each control option. The user exhibited superior device controllability across multiple limb positions using our regression-based solution. This work contributes evidence that a deep learning regression control approach can elicit accurate, simultaneous, and proportional device movements, while mitigating the limb position effect for a transradial prosthesis user.