A myoelectric digital twin for fast and realistic modelling in deep learning

Area of Science:

• Biomedical Engineering
• Machine Learning
• Human-Machine Interfaces

Background:

• Muscle electrophysiology is crucial for human-machine interfaces (HMIs) in robotics and virtual reality.
• Current decoding algorithms lack the sophistication for fine control required by advanced HMIs.
• Deep learning shows promise but needs large, high-quality annotated datasets, which are difficult to obtain for electromyography (EMG).

Purpose of the Study:

• To introduce a novel, computationally efficient simulation model for generating realistic electromyography (EMG) data.
• To enable the creation of large, perfectly annotated datasets for training deep learning algorithms.
• To overcome the data acquisition bottleneck in EMG-based HMI development.

Main Methods:

• Development of a "Myoelectric Digital Twin" - a fast, realistic computational model for EMG signal simulation.
• Utilizing the Myoelectric Digital Twin to generate synthetic, annotated EMG datasets.
• Applying these simulated datasets to train deep learning algorithms for muscular signal decoding.

Main Results:

• The Myoelectric Digital Twin provides a computationally efficient method for simulating realistic EMG signals.
• It enables the generation of arbitrarily large, perfectly annotated datasets.
• This approach facilitates the training of deep learning models for improved EMG signal decoding.

Conclusions:

• The Myoelectric Digital Twin concept addresses the need for efficient data generation in EMG-based deep learning.
• It significantly accelerates the development cycle for sophisticated human-machine interfaces.
• This innovation opens new avenues for muscular signal decoding and HMI applications.