Counteracting Electrode Shifts in Upper-Limb Prosthesis Control via Transfer Learning

Area of Science:

• Biomedical Engineering
• Machine Learning
• Rehabilitation Robotics

Background:

• Machine learning (ML) shows promise for upper-limb prosthesis control.
• Real-world application is hindered by disturbances like electrode shifts.
• Adaptive transfer learning offers a potential solution to minimize retraining data needs.

Purpose of the Study:

• To present a novel, simplified transfer learning method for prosthesis control.
• To conduct the first user study evaluating transfer learning's efficacy against electrode shifts.
• To introduce the Box and Beans test for assessing prosthesis proficiency.

Main Methods:

• Implemented a simple transfer learning technique.
• Designed and utilized the Box and Beans test for user evaluation.
• Compared performance across three conditions: initial system, system with electrode shifts, and system after transfer learning.

Main Results:

• Transfer learning significantly mitigated the negative impact of electrode shifts.
• User performance in the Box and Beans test improved after applying transfer learning.
• Demonstrated the practical effectiveness of the proposed transfer learning approach.

Conclusions:

• The novel transfer learning method is effective in counteracting electrode shifts.
• This approach enhances the robustness of ML-based prosthesis control for daily use.
• Transfer learning represents a viable strategy for improving prosthetic device reliability.