Prosthetic Control Using Multiple Degrees of Freedom of the Shoulder as Inputs and ANNS

Abstract

The use of myoelectric signals (MES) for prosthetic control has been in continuous development at the Institute of Biomedical Engineering (IBME) for many years. Although this control scheme has shown to be successful for upper extremity prostheses, it offers no real feedback mechanism which can provide some level of
cognitive awareness of the prostheses’ overall performance. D. C. Simpson (1974) developed a control principle which reduces the conscious mental effort demanded while relaying useful feedback information to the user. The principle, which is known as extended physiological proprioception (EPP), uses the position of intact joints as an input signal for prosthetic control.

The shoulder joint motions are of particular interest as input signals since they are the foundation upon which most arm motions are initially based and the prosthesis acts as an extension of the residual limb of an amputee. We proposed to use three degrees of freedom of the shoulder (using fibre optic sensors) as possible control inputs for the artificial neural network control system of an above-elbow prosthesis. The goal was to develop artificial neural networks capable of mapping the trajectory of the prosthesis joints for a number of primitive motions found in activities of daily living. With the shoulder joint information available, the control inputs contain much more information and would improve the range of motion accepted by the controller. This should also increase the accuracy of the predicted prosthesis joint positions. Preliminary results indicate that a time delay neural network is a robust architecture for this application.