Directional Forgetting for Stable Co-Adaptation in Myoelectric Control

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Human-Computer Interaction

Background:

• Myoelectric controllers map electromyographic signals to prosthetic functions.
• Conventional systems require extensive user training due to performance instabilities.
• Co-adaptation allows concurrent user learning and machine adaptation but depends on data weighting.

Purpose of the Study:

• To propose a novel directional forgetting approach for recursive least-squares algorithms in myoelectric control.
• To improve the efficiency and stability of prosthetic function adaptation.
• To reduce the sensitivity of adaptation algorithms to parameter selection.

Main Methods:

• Developed a function-driven directional forgetting algorithm for recursive least-squares.
• Implemented a scheme that discounts past data only in the direction of new data.
• Validated the approach through real-time myoelectric tasks with human subjects.

Main Results:

• Directional forgetting with a 0.995 forgetting factor significantly outperformed exponential forgetting and unassisted learning.
• Myoelectric control demonstrated stable adaptation across various forgetting factors using directional forgetting.
• The proposed method minimized distortion in other data regions during local corrections.

Conclusions:

• Directional forgetting offers a more effective strategy for adapting myoelectric control systems.
• This approach enhances prosthetic performance and alleviates parameter selection sensitivities.
• Selective data discounting improves the robustness and user experience of myoelectric prosthetics.