Adaptive Filter for Biosignal-Driven Force Controls Preserves Predictive Powers of sEMG

Area of Science:

• Biomedical Engineering
• Machine Learning
• Signal Processing

Background:

• Machine learning-based electromyography (EMG) controls require stable and repeatable muscle activity signals.
• Signal instability and delays limit EMG applications like exoskeleton control and teleimpedance, exacerbated by factors like gravity compensation and signal heteroscedasticity.

Purpose of the Study:

• To introduce and characterize an adaptive filter for surface EMG (sEMG) features.
• To improve the stability and responsiveness of sEMG signals for applications with low delay tolerances.

Main Methods:

• An adaptive filter was developed to automatically adjust its cutoff frequency based on movement intention.
• The filter was tested offline and online using a regression-based joint torque predictor.

Main Results:

• The adaptive filter demonstrated more accurate predictions, indicated by a lower root mean square error compared to unfiltered predictions.
• The filter provided higher signal responsiveness with reduced lag compared to conventional low-pass filters.

Conclusions:

• The proposed adaptive filter enhances the performance of EMG-based control systems by improving signal accuracy and responsiveness.
• This adaptive filtering approach offers a viable solution for applications demanding precise and timely control, such as advanced prosthetics and robotics.