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

Older adults frequently face difficulties in activities of daily living (ADLs) due to age-related declines in strength, coordination, and perception. Myoelectric control provides an intuitive human-robot interface by translating muscle activity into assistive commands. However, its practical application is still challenged by signal annotation, multijoint coordination, and cross-task generalization. This study proposes a 3-level intelligent framework for multijoint upper-limb assistance based on electromyography (EMG) to support the daily living activities of older adults. At the physiological level, situation-aware labeling protocols matched to different EMG conditions are proposed to reduce annotation ambiguity and improve robustness to signal changes. At the functional level, focusing on elemental joint activities, a deep backbone model is designed to infer both single-joint movements and coordinated multijoint patterns with an accuracy of 95.34%. At the behavioral level, the model is further distilled to support complex ADL tasks with human-robot interactions while continually incorporating new knowledge without catastrophic forgetting. The framework is implemented in real time on an EMG-controlled multijoint robotic system, providing smooth and coordinated assistance in daily activities. Overall, the proposed framework provides a systematic solution for EMG-based multijoint coordination, encompassing the entire pathway from physiological signal processing to functional intent decoding and behavioral adaptation during daily activities. It offers a technical approach to coordinated upper-limb assistance and lays a broader foundation for the design of practical and adaptive assistive systems, contributing to improved autonomy for older adults and supporting the broader societal goal of healthy aging.