Samueli School of Engineering

Traditional prosthetics are often prohibitively expensive ($5,000–$100,000+) and can require invasive medical procedures to function. To address this, we developed a low-cost, electromyography (EMG) controlled prosthetic hand that utilizes an embedded convolutional neural network (CNN) to translate muscle signals into mechanical motion. Using a non-invasive dry-electrode on the wrist, raw EMG data is processed and classified in under 40 milliseconds on average. The CNN accurately identifies three predefined hand gestures with >90% accuracy. By keeping total manufacturing costs, including electronics, mechanical hardware, and filament, to just $260, this project demonstrates the viability of highly accessible, neural-network-driven prosthetics.