Towards the Myoelectric Digital Twin: Ultra Fast and Realistic Modelling for Deep Learning

Abstract

Muscle electrophysiology has emerged as a powerful tool to drive human machine interfaces, with many new recent applications outside the traditional clinical domains. It is currently a crucial component of control systems in robotics and virtual reality. However, more sophisticated, functional, and robust decoding algorithms are required to meet the fine control requirements of these new applications. Deep learning approaches have shown the highest potential in this regard. To be effective, deep learning requires a large amount of high-quality annotated data for training; the only option today is the use of experimental electromyography data. Yet the acquisition and labelling of training data is time-consuming and expensive. Moreover, the high-quality annotation of this data is often not possible because the ground truth labels are hidden. Data augmentation using simulations, a strategy applied in other deep learning applications, has never been attempted in electromyography due to the absence of computationally efficient and realistic models. Here, we present a new highly realistic and ultra-fast computational model tailored for the training of deep learning algorithms. For the first time, we are able to simulate arbitrary large datasets of realistic electromyography signals with high internal variability and leverage it to train deep learning algorithms. Because the computational model provides access to all the hidden parameters of the simulation, it also allows us to use some annotation strategies that are impossible with experimental data. We believe that this concept of Myoelectric Digital Twin allows new unprecedented approaches to muscular signals decoding and will accelerate the development of human-machine interfaces.