UKRI AI CDT in Dependable and Deployable AI for Robotics (CDT-D2AIR)

Ada Grecner

Project
Deployable Machine Learning sEMG Signal Classification for Low-Power Upper-Limb Prosthetic Controllers

Summary
State-of-the-art myoelectric upper limb prosthetics are abandoned by their users at a non-negligible rate. The abandonment issues have been well documented, and yet commercial devices from industry have so far failed to resolve them. We propose utilising adaptive learning on deep networks, combined with user-first robust design, to deploy a prosthetic controller system on low-powered microcontrollers through TinyML. We hope that the main contribution of our will be expanding the domain of ML on edge devices, as well as serve as a proof of concept of a real-world ready EMG-based human-machine interface, tackling the many bugbears that lead to prosthetic abandonment.

Lead Institution
School of Mathematical and Computer Sciences, Heriot-Watt University

Supervisors
Marta Vallejo (Heriot-Watt University), Rob Stewart (Heriot-Watt University)