Bio-robotics and rehabilitation engineering: a reproducible framework for EMG-driven prosthetic control

Abstract

The integration of bio-robotics and rehabilitation engineering is reshaping healthcare by providing assistive technologies that restore mobility and independence to individuals with neuromuscular impairments, stroke, or limb loss. Among the many approaches, electromyography (EMG)-driven prosthetic control has emerged as a powerful technique for decoding user intent and actuating robotic devices. However, challenges such as noisy biosignals, dataset limitations, and reproducibility hinder progress and slow clinical translation. In this study, we present a reproducible computational framework for EMG-based prosthetic hand control using synthetic data. EMG signals representing grip and relax states were generated to simulate muscle activation and baseline rest. Signals were filtered within the 20-450 Hz EMG band, and three physiologically meaningful features root mean square (RMS), variance, and waveform length were extracted. A support vector machine (SVM) with a radial basis function kernel classified the signals with perfect accuracy, achieving 100% precision, recall, and F1-scores across both classes. Representative plots, classification tables, and confusion matrix analysis confirmed the separability of synthetic grip and relax states. The proposed framework contributes to rehabilitation engineering by providing an accessible, reproducible pipeline that enables rapid prototyping of EMG-based algorithms before applying them to real-world datasets such as NinaPro. This approach lowers the barrier to entry, accelerates research, and demonstrates how lightweight computational methods can be adapted for real-time prosthetic and rehabilitation robotics applications.