Sensor-Agnostic, LSTM-Based Human Motion Prediction Using sEMG Data.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Neuroscience

Background:

• Surface electromyography (sEMG) is widely used for motion classification and prediction.
• Limitations exist due to variations in sEMG sensor hardware.
• Deep learning approaches are increasingly popular for motion prediction.

Purpose of the Study:

• To investigate the impact of different sEMG sensor hardware platforms on deep learning model performance.
• To evaluate the ability of a neural network to predict arm angular trajectory using data from diverse sEMG sensors.

Main Methods:

• Collected raw sEMG data from subjects performing exercises using two different sensor platforms.
• Trained a bidirectional long short-term memory (bi-LSTM) neural network to predict one-degree-of-freedom (DoF) angular trajectory.
• Analyzed the influence of sensor configurations including communication, DAQ, electrodes, buffering, preprocessing, and sampling frequency.

Main Results:

• Deep learning neural networks trained on sEMG data from different hardware platforms exhibited similar performance.
• The bi-LSTM networks demonstrated consistent predictive capabilities regardless of sensor origin.
• This suggests that the neural network architecture is resilient to variations in data acquisition hardware.

Conclusions:

• Deep learning models for sEMG-based motion prediction demonstrate hardware-agnostic characteristics.
• The findings support the use of deep learning for reliable motion prediction across different sEMG sensor systems.
• Future applications can leverage these robust models without being constrained by specific hardware choices.