An Adaptive Multi-Modal Control Strategy to Attenuate the Limb Position Effect in Myoelectric Pattern Recognition

Area of Science:

• Biomedical Engineering
• Robotics
• Machine Learning

Background:

• Pattern recognition algorithms are crucial for myoelectric control of upper limb prostheses.
• Current methods struggle with real-world variations like limb position and external loads.
• There is a need for adaptive control strategies in prosthetic devices.

Purpose of the Study:

• To propose and evaluate an adaptive pattern recognition method for myoelectric control.
• To enhance prosthesis control by adapting to underrepresented variations in limb position and external loads.
• To improve the robustness and performance of myoelectric control systems.

Main Methods:

• Developed an adaptive pattern recognition classifier utilizing an augmented dataset.
• The dataset included variations in limb position and external loads.
• Evaluated the method through target achievement control tests with ten able-bodied volunteers.

Main Results:

• The adaptive algorithm demonstrated a higher median completion rate (>3.33%) compared to the baseline classifier.
• Subject-specific analysis indicated potential for improved control after adaptation (≤13% completion rate).
• Adapted points provided new information within classes, suggesting enhanced adaptability.

Conclusions:

• The proposed adaptive method shows significant potential for improving myoelectric control of upper limb prostheses.
• Adaptation to variations in limb position and external loads enhances classifier performance.
• Further development is encouraged based on promising preliminary results.