Elbow movement estimation based on EMG with NARX Neural Networks.

Area of Science:

• Biomedical Engineering
• Robotics
• Neuroscience

Background:

• Surface electromyography (sEMG) offers a natural interface for controlling robotic systems like exoskeletons.
• Key challenges include human motor redundancy and sEMG signal variability, hindering precise control.
• Accurate estimation of limb trajectory from sEMG is crucial for seamless human-robot interaction.

Purpose of the Study:

• To develop and validate a feature extraction and classification method for accurate elbow angular trajectory estimation using sEMG.
• To implement a Nonlinear Auto Regressive with Exogenous inputs (NARX) Neural Network for real-time control feasibility.
• To assess the potential for controlling more complex movements, including shoulder joints.

Main Methods:

• sEMG data from Biceps and Triceps Brachii muscles were collected during an elbow flexo-extension task.
• Pre-processed sEMG signals were divided into five frequency intervals.
• These frequency intervals were used as input features for a NARX Neural Network.

Main Results:

• The NARX Neural Network achieved a high correlation between estimated and measured elbow angular trajectories.
• The root-mean-square error (RMSE) was a maximum of 7 degrees.
• The developed procedure demonstrated feasibility for real-time implementation after model training.

Conclusions:

• The proposed sEMG processing and NARX network approach provides accurate elbow trajectory estimation.
• This method shows promise for real-time control of exoskeletons and other robotic devices.
• Future work can extend this approach to more complex movements and joints, such as the shoulder.