Development and electromyographic control of a bionic robotic arm via a neural network for gesture recognition

Abstract

This diploma thesis focuses on the development of a bionic robotic arm. Its operation is controlled by an artificial intelligence model designed to classify surface electromyographic (sEMG) signals. The prosthetic limb is primarily composed of 3D printed components and servomotors responsible for actuating individual hand digits. This robotic system mirrors the user’s hand gestures in real-time by acquiring, processing and classifying electromyographic data. These data are gathered from three sensors which are placed on the forearm muscles located near the elbow, using electrodes. The proposed classifier is a multilayer perceptron model, which was trained on feature vectors extracted from data segments. To evaluate the efficacy of this classifier, the segment length and overlap size were altered and each time the prediction accuracy was examined by computing key metrics such as the F1 score and confusion matrix. This diploma thesis concludes with a series of observations on the system’s behavior, along with methods that could be implemented in the future to further refine and optimize the proposed electromyographic data-driven controller.