Improving Pattern Recognition based Myocontrol of Prosthetic Hands via User-in-the-Loop

Abstract

The control of prosthetic hands based on electromyography (EMG) has the potential to restore motor functions to patients who have undergone an amputation, significantly improving their quality of life. However, there are unresolved issues in achieving control that is rich in movements and stable, including the presence of the limb position effect. The thesis focuses on trying to reduce this cause of instability by proposing a modified version of the Ridge Regression with Random Fourier Features algorithm, made incremental and enriched with user feedback. This approach is then validated through an experiment on 12 non-disabled subjects to verify the increase in performance, and through a further pilot study on an subject with amputation, following the adaptation of the software to a prosthetic hand under development.