Pattern Recognition Systems for Myoelectric Control of Biomechatronic Prosthetic Hands

The human hand is considered as the highest example of dexterous system capable of interacting with different objects adapting its manipulation abilities to them. Therefore, the hand loss causes severe impairment for the amputee and can significantly reduce quality of life. In the last 70 years there have been significant improvements in the upper limb prosthetic hand field thanks to the advancements in the technological field and in the surgical procedure leading to prosthetic devices that are more conceived to reproduce aesthetical as well as functional features of the lost limb. However, the currently adopted hand prosthesis surface electromyography (sEMG) control strategies, representing the clinical state of art, do not provide the users with a natural control feeling and do not exploit all the potential of commercially available multi-fingered hand prostheses. Pattern recognition (PR) and machine learning techniques applied to sEMG can be effective for a natural control based on the residual muscles contraction of amputated people corresponding to phantom limb movements. As the researches has reached an advanced grade accuracy, these algorithms are mature and widely validated, and the embedding is necessary for the realization of prosthetic devices. This thesis wants to address the specific issue of enhancing both the performance and the control feeling of existing multi-grasp prosthetic hands, by designing a new embedded control based on pattern recognition algorithms applied to sEMG signals. To this purpose, firstly a comparison among different supervised machine learning techniques on data collected from 30 people with trans-radial amputation is carried out in order to provide innovative engineering tools and indications on how to choose the most suitable classification algorithm based on the application and the desired results for prostheses control. Then, the obtained result has been used for the design and evaluation of an embedded control system (hardware-firmware-software) for hand prostheses capable to handle up to five different grasping movements, successfully tested on amputee subjects. As complementary activity, this thesis proposes a new approach for neural control of hand prostheses, grounded on pattern recognition applied to the envelope of neural signals. It was demonstrated that it is possible to apply the well-known techniques of EMG pattern recognition to a conveniently processed neural signal and can pave the way to the application of neural gesture decoding in upper limb prosthetics. This intends to overcome limitations of traditionally adopted techniques of sEMG and ENG processing allowing both to control a prosthetic device and to stimulate the PNS.