Multi-Grip Classification-Based Prosthesis Control With Two EMG-IMU Sensors.

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Machine Learning in Prosthetics

Background:

• Myoelectric control of upper-limb prostheses has long proposed machine learning for intuitive grip selection.
• Commercial adoption of these systems is increasing, but often requires numerous electrodes.
• A need exists for simplified, yet effective, control strategies.

Purpose of the Study:

• To propose an end-to-end strategy for multi-grip, classification-based prosthesis control using only two sensors.
• To minimize unintended prosthesis activations by accurately estimating probabilities and rejecting low-confidence predictions.
• To evaluate the efficacy of a confidence-based error rejection strategy.

Main Methods:

• Utilized electromyography (EMG) electrodes and inertial measurement units (IMUs) as the sole sensors.
• Developed a classification-based control strategy for multi-grip selection.
• Implemented a confidence-based error rejection strategy with grip-specific thresholds.

Main Results:

• The proposed system demonstrated effective real-time pick and place capabilities.
• Evaluated with 12 able-bodied and 2 transradial amputee participants.
• Results indicate the potential for intuitive, classification-based multi-grip control with minimal system modifications.

Conclusions:

• A two-sensor system (EMG and IMUs) can achieve effective multi-grip prosthesis control.
• Confidence-based error rejection is crucial for minimizing unintended activations.
• The proposed strategy shows promise for enhancing existing upper-limb prosthetic systems.