Myoelectric Control based on Machine Learning of a Low-Cost 7-DOF Transhumeral Prosthetic Arm through TinyML

Abstract

Amputation of the upper limbs is still a major and disabling problem, especially in the low-resource setting environment in which the price of cutting-edge commercial prosthetics (which can often cost USD 70,000 or more) is highly prohibitive. This study offers a comprehensive technical paradigm of development, optimization, and testing of a 7-degree of freedom (DOF) transhumeral prosthetic arm driven by a power-source-embedded artificial intelligence (TinyML). This is achieved by using a computationally paired ESP32-S3 microcontroller; after pruning the Random Forest (RF) classifier, we obtain an accuracy of 92.3% in gesture recognition and an end-to-end deterministic latency of 88ms. This paper describes the hardware-software co-design, the mathematical process modeling of surface Electromyography (sEMG) signal processing with digital Z-transforms, and a new 60-second incremental learning protocol suitable for quick subject-specific calibration. Moreover, the system has been combined with a dual-rail power distribution network and an 8-channel analog multiplexing approach to address hardware constraints of low-cost microcontrollers. With a total bill of materials (BOM) of approximately PKR 60,000-80,000(USD 215-286), this study is a representation of a research paradigm shift in valuebased, affordable assistive technology for humanitarian deployment.