Control of prosthetic hands is still an open problem, currently, commercial prostheses use direct myoelectric control for this purpose. However, as mechanical design advances, more dexterous prostheses with more degrees of freedom (DOF) are created, then a more precise control is required. State of the art has focused in the use of pattern recognition as a control strategy with promising results. Studies have shown similar results to classic control strategies with the advantage of being more intuitive for the user. Many works have tried to find the algorithms that best follows the user’s intention. However, deployment of these algorithms for real-time classification in a prosthesis has not been widely explored. This paper addresses this problem by deploying and testing in real-time an Artificial Neural Network (ANN). The ANN was trained to classify three different motions: no grasp, precision grasp and power grasp in order to control a two DOF trans-radial prosthetic hand with electromyographic signals acquired from two channels. Static and dynamic tests were made to evaluate the ANN under those conditions, 95% and 81% accuracy scores were reached respectively. Our work shows the potential of pattern recognition algorithms to be deployed in microcontrollers that can fit inside myoelectric prostheses.

This content is only available via PDF.
You do not currently have access to this content.