SEMG-Based Prosthetic Hand with an Integrated Mobile Application

The main focus of this study was to design and develop an assistive device in the form of a low-cost prototype prosthetic hand that utilizes cords as tendons, inexpensive servos, and a simplified actuation method. Additionally, the objective was to create a neural network using only two surface electromyography (sEMG) sensors. Various methods were experimented with to achieve the desired outcome, including transfer learning with RESNET50 and the implementation of a Convolutional Neural Network. The UCI-sEMG dataset was used to train the proposed model for hand movements. Furthermore, a mobile application was developed to control the prosthetic hand, aiming to enhance interactivity and expand the range of control. The entire system communicates through gRPC. The results demonstrate the feasibility of creating a prosthetic hand that internally stores all the cords and servos, while still providing a satisfactory range of motion for each finger. The evaluation process also indicates the algorithm’s ability to predict hand gestures, although its performance may not be on par with state-of-the-art models. Nevertheless, considering the limited number of sensors used, the achieved accuracy of \(58\%\) shows promise. Additionally, the mobile application features a simple yet functional design that meets the requirements for support, ease of use, and flexibility.