The project focuses on improving motion-intention recognition for assistive robotic systems, particularly robotic hand prostheses for transradial amputees. It uses surface electromyography signals from residual muscles to classify different hand gestures, allowing the prosthesis or assistive robot to better interpret the user’s intended movement.

The study systematically compares several machine-learning and deep-learning models for hand-gesture classification. It evaluates how different preprocessing choices, time-domain features, and sliding-window lengths affect recognition accuracy and generalization across gesture classes and users.

The results show that ensemble learning and deep-learning methods generally performed better than classical machine-learning approaches. Overall, the project provides useful guidance for developing more reliable, responsive, and user-adaptive EMG-based control systems for prosthetic hands and other assistive robots.