Deep Learning for Arm Movement Simulation With EMG Signals

Abstract

This paper presents a real-time electromyography (EMG)-based arm movement detection system using deep learning for prosthetics and rehabilitation. Dual-channel EMG sensors capture grip and release actions, processed through hardware-level thresholding. A CNN-BiLSTM model classifies the signals, and realtime hand animation is rendered via MATLAB SynGrasp. Thresholding was chosen over raw analog processing due to noise, grounding issues, and signal instability. The system demonstrates high classification accuracy, low latency, and robust real-world performance.