Hybrid CNN-LSTM Deep Learning Framework for EMG-Based Control of Bionic Prosthetic Limbs

Abstract

The design and development of functional, affordable prosthetic limbs continue to present a maj or global challenge, particularly in developing regions where limb loss due to conflict or disease is prevalent. This study proposes a cost-effective bionic arm system driven by electromyography (EMG) signals and optimized through deep learning (DL) techniques to achieve high precision and naturalistic motion. Drawing inspiration from the open-source InMoov humanoid robot platform, the project aims to provide an accessible solution for individuals with limb loss resulting from trauma or medical conditions. The proposed methodology involves the acquisition of EMG signals from users, followed by comprehensive preprocessing, comprising filtering, rectification, scaling, encoding, and subsequent classification using a hybrid DL model that integrates a convolutional neural network (CNN) with long short-term memory (LSTM) networks. The developed system achieved a gesture recognition accuracy of 97%, successfully translating neural intent into precise mechanical actuation of a custom-designed bionic arm. The results demonstrate the feasibility of combining low-cost hardware with advanced DL models to create practical and efficient assistive technologies. This research lays the foundation for future developments focused on enhancing system responsiveness, accuracy, and adaptability through wireless and real-time control mechanisms.