Artificial-Intelligence-Driven Electromyography Adaptation for Elderly Assistance at Physiological, Functional, and...

Area of Science:

• Biomedical Engineering
• Rehabilitation Robotics
• Human-Robot Interaction

Background:

• Older adults experience functional decline impacting Activities of Daily Living (ADLs).
• Myoelectric control (electromyography - EMG) offers intuitive human-robot interfaces but faces challenges in signal processing and coordination.
• Existing systems struggle with robust signal annotation, multijoint control, and task generalization.

Purpose of the Study:

• To develop and validate a 3-level intelligent framework for EMG-based multijoint upper-limb assistance tailored for older adults.
• To improve the accuracy and robustness of myoelectric control for assistive robotic systems.
• To enhance the autonomy and quality of life for older adults through adaptive robotic support.

Main Methods:

• A 3-level framework integrating physiological signal processing, functional intent decoding, and behavioral adaptation.
• Situation-aware labeling protocols for improved EMG signal robustness.
• A deep backbone model for inferring single-joint and multijoint movements with high accuracy.
• Model distillation for adapting to complex ADL tasks and continuous learning without catastrophic forgetting.

Main Results:

• The deep backbone model achieved 95.34% accuracy in inferring joint movements.
• The framework was successfully implemented in real-time on an EMG-controlled robotic system.
• The system provided smooth and coordinated assistance during daily activities.

Conclusions:

• The proposed framework offers a systematic solution for EMG-based multijoint coordination in assistive robotics.
• It bridges the gap from physiological signal processing to adaptive behavioral control.
• This work provides a foundation for practical, adaptive assistive systems, promoting independence for older adults and supporting healthy aging.