“Myocontrol of Prosthetic Hands: Enforcing Active Learning by a Machine Learning based Fault Detector”

Improvements in the control of prosthetic hands can provide a future solution for amputated people, improving their motor skills and in general their quality of life. The prosthesis controlled by electromyography (EMG) are studied for their greater potential, but the technology is still unprepared for mass usage. The control is too unstable due to various phenomena related to the inconstancy of bioelectric signals and changes in the human body posture during the execution of tasks. This thesis is focused on the creation of a fault detector able to detect these instabilities and correct the prosthesis’s future behaviour. Machine learning algorithms and methodologies were employed for this objective. The approach is then tested on four healthy subjects with a preliminary experiment.