Limb position effect in myoelectric control: strategies for optimisation and standardisation

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Human-Machine Interfaces

Background:

• Myoelectric control, utilizing electromyography (EMG) signals, is crucial for advanced prosthetics, AR/VR, and consumer electronics.
• Limb position variability during daily activities significantly degrades the reliability of myoelectric controllers.
• Developing robust myoelectric control requires mitigating the limb position effect.

Purpose of the Study:

• To investigate methods for reducing the limb position effect in myoelectric control.
• To assess the impact of training data volume and modality on classifier accuracy.
• To introduce an open-source device for standardizing myoelectric control experiments.

Main Methods:

• An open-source, automated device with 16 arm positioning locations was developed to standardize myoelectric control experiments.
• Data from 19 participants, including one with a congenital upper-limb difference, were collected across static and dynamic limb conditions.
• Forearm EMG and upper limb kinematics were recorded, and linear discriminant analysis models were trained and evaluated.

Main Results:

• Classification accuracy decreased with untrained limb positions, confirming the limb position effect.
• Training with four positions optimized accuracy while balancing training burden.
• Multi-modal classifiers integrating EMG and kinematic data achieved optimal performance when trained with dynamic limb data.

Conclusions:

• The limb position effect in myoelectric control can be effectively countered by training with multiple limb positions and incorporating kinematic data.
• Multi-modal classifiers trained with dynamic limb data offer high accuracy with minimal training burden.
• The study presents novel findings for congenital limb differences and introduces a standardized open-source device to advance myoelectric control research.