Simultaneous Three-Degrees-of-Freedom Prosthetic Control Based on Linear Regression and Closed-Loop Training...

Area of Science:

• Biomedical Engineering
• Rehabilitation Robotics
• Machine Learning in Healthcare

Background:

• Machine learning controllers for prostheses using electromyographic (EMG) signals are increasingly popular.
• Regression-based control offers more natural, proportional movement than discrete classification methods.
• Existing regression controllers rarely manage more than two degrees of freedom simultaneously.

Purpose of the Study:

• To apply an adaptive linear regressor for simultaneous, proportional control of three degrees of freedom (DoF) in prosthetic hands.
• To investigate the impact of training paradigms on controller performance.
• To introduce a closed-loop training procedure for improved EMG signal generation and controller learning.

Main Methods:

• Utilized an adaptive linear regressor with eight EMG sensors for a low-dimensional feature space.
• Implemented a novel closed-loop training procedure where users actively improved EMG signal quality.
• Tested the system on 10 healthy and 3 limb-deficient subjects for simultaneous control of left-right, up-down, and open-close hand movements.

Main Results:

• The proposed closed-loop training procedure significantly improved controller performance.
• The average completion rate for simultaneously controlling three DoF increased from 53% to 65%.
• The combination of multidimensional targets and the training protocol was key to performance enhancement.

Conclusions:

• A closed-loop training paradigm is crucial for optimizing machine learning-based EMG controllers for prostheses.
• Effective simultaneous control of multiple DoF is achievable with adaptive linear regression and improved training.
• This approach holds promise for more natural and intuitive prosthetic limb control.