A Learning Scheme for EMG Based Decoding of Dexterous, In-Hand Manipulation Motions

Area of Science:

• Biomedical Engineering
• Robotics
• Human-Computer Interaction

Background:

• Electromyography (EMG) is crucial for controlling assistive devices, but decoding continuous human motion remains challenging.
• Existing research primarily focuses on decoding human intention rather than continuous motion.
• Dexterous in-hand manipulation tasks require precise control, highlighting the need for advanced EMG decoding methods.

Purpose of the Study:

• To develop and evaluate a learning scheme for decoding object motion during dexterous in-hand manipulation using EMG signals.
• To investigate the influence of different muscles, gender, and hand size on EMG-based motion decoding accuracy.
• To compare the performance of subject-specific, hand-specific, and object-specific decoding models against generic models.

Main Methods:

• Utilized EMG signals from 16 muscle sites (hand and forearm) of 11 subjects.
• Employed an optical motion capture system to record object motion.
• Formulated object motion decoding as a regression problem using the Random Forests methodology with time-domain features (RMS, waveform length, zero crossings).
• Implemented a 10-fold cross-validation for model assessment and calculated feature importance.

Main Results:

• Subject-specific, hand-specific, and object-specific decoding models demonstrated superior accuracy compared to generic models.
• Identified the contribution of different muscles to decoding accuracy.
• Analyzed the impact of gender and hand size on the overall decoding performance.

Conclusions:

• Personalized EMG decoding models (subject, hand, or object-specific) are more effective for continuous motion decoding in in-hand manipulation tasks.
• Understanding muscle contributions and individual variations (gender, hand size) can further optimize EMG-based control systems.
• This approach advances the development of more intuitive and accurate control for robotic, prosthetic, and assistive devices.