Hierarchical myoelectric control of a human upper limb prosthesis

Myolectric control is nowadays the most used approach for electrically-powered upper limb prostheses. The myoelectric controllers use electromyographic (EMG) signals as inputs. These signals can be collected from the skin surface using surface EMG sensors, or intramuscular, using needle sensors. No matter which method is used, they have to be processed before being used as controller inputs. In this paper, we present an algorithm based on an autoregressive (AR) model representation and a neural network, for EMG signal classification. The results have shown that combining a low-order AR model with a feed-forward neural network, a rate of classification of 98% can be achieved, while keeping the computational cost low. We also present a hierarchical control architecture and the implementation of the high-level controller using Finite State Machine. The solution proposed is capable of controlling three joints (i.e. six movements) of the upper limb prosthesis. The inputs of the high-level controller are obtained from the classifier, while its outputs are applied as input signals for the low-level controller. The main advantage of the proposed strategy is the reduced effort required to the patient for controlling the prosthetic device, since he only has to initiate the movement that is finalized by the low-level part of the controller.