Stable myoelectric control of a hand prosthesis using non-linear incremental learning.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Rehabilitation Robotics

Background:

• Stable myoelectric control of hand prostheses is challenging due to signal variability.
• Current machine learning methods require frequent retraining, limiting real-world application.
• Surface electromyography (sEMG) offers limited degrees of freedom for prosthesis control.

Purpose of the Study:

• To develop a non-linear incremental learning method for adaptive myoelectric control.
• To enable continuous adaptation to changes in myoelectric signals.
• To improve the stability and reliability of prosthetic hand control.

Main Methods:

• Utilized Incremental Ridge Regression combined with Random Fourier Features.
• Predicted finger forces from surface electromyography signals.
• Collected sEMG and force data from 10 subjects over multiple sessions.
• Implemented an on-line teleoperation system with a prosthetic hand.

Main Results:

• Demonstrated effective adaptation with small, incremental updates to the learning model.
• Showcased stable performance in predicting single-finger forces.
• Enabled reliable grasping, carrying, and releasing of objects using a prosthetic hand.
• Achieved stable grasping despite signal changes and arm/wrist movements.

Conclusions:

• The proposed incremental learning method effectively addresses the challenge of myoelectric signal variability.
• This approach offers a practical and stable solution for continuous prosthetic hand control.
• The findings support the potential for improved human-machine interfaces in prosthetics.