Key points are not available for this paper at this time.
Machine learning (ML) technology enables the decoding of surface electromyographic (sEMG) signals for prosthetic control. However, its deployment beyond desktop studies remains challenging. A key barrier is the lack of computing resources on battery-operated embedded systems. In this paper, we demonstrate the design and development of an embedded platform which applied a Tensorflow Lite ML model for prosthetic control for the first time. We optimised the feature extraction stage as well as the model design so that they could be executed by a low-power microcontroller in real-time. The testing results with able-bodied participants showed that the online performance of the classification model on our system is similar to the offline testing accuracy on a PC but with lower memory consumption and computational power. With this paper, we aim to demonstrate the value and the potential of adopting TinyML and low-cost embedded systems in accelerating out-of-laboratory research on limb prosthetics as well as pre-clinical studies.
Wu et al. (Thu,) studied this question.