As people age, weakening muscle strength, coordination, and sensory feedback can make everyday tasks—such as dressing, carrying objects, cooking, and using tools—harder to perform independently. Assistive technologies have progressed, but a core bottleneck remains: robots must reliably interpret human motor intent in real time. Electromyography (EMG) offers a direct window into neuromuscular control, appearing before visible movement and enabling a natural interface for assistive robotic systems.
Yet deploying EMG control in daily life is challenging. Reliable intent decoding depends on accurate signal labeling, while manual annotation and external motion-capture setups add cost and complexity. Many existing approaches also concentrate on single-joint recognition, limiting their ability to interpret coordinated multi-joint actions such as combined hand–elbow movements. Finally, EMG signals shift with task context, posture, and scenario, creating generalization failures and potential “forgetting” of previously learned capabilities.
To address these issues, researchers at City University of Hong Kong developed a three-level, AI-driven EMG adaptation framework for upper-limb assistance in older adults. The system translates EMG into robotic actions using physiological, functional, and behavioral layers. Upper-limb EMG data were collected from both younger and older participants, using four muscles associated with hand and elbow motion: triceps, biceps, extensor digitorum, and flexor digitorum superficialis.
At the physiological level, the team introduced selective active labeling and contextual labeling strategies tailored to different activation patterns, reducing reliance on labor-intensive manual annotation and external sensing. At the functional level, a one-dimensional convolutional neural network predicts both hand states (relaxed, open, closed) and elbow states (relaxed, flexed, extended) from 4-channel EMG. A voting mechanism stabilizes real-time outputs for consistent control.
At the behavioral level, knowledge distillation enables the model to learn new daily tasks—such as cooking, lifting a bag, and pulling a grocery cart—while preserving basic joint-movement knowledge. The framework was integrated into a real-time robotic control setup using EMG acquisition, edge computing, and a 6-axis collaborative robotic arm for validation in everyday assistance scenarios.
Results demonstrated that the three-level approach improved EMG labeling quality, multi-joint intent recognition, and task adaptation. Selective labeling fit hand opening and closing due to strong, brief EMG bursts, while contextual labeling better matched the more continuous EMG patterns seen during elbow flexion and extension. Offline prediction accuracy reached 95.42% for hand states and 93.97% for elbow states.
In coordinated real-time control, the system achieved 95.34% overall accuracy across nine hand–elbow combinations and drove smooth robotic assistance. During the cooking scenario, knowledge distillation notably boosted performance for a student model relative to its teacher, improving hand–elbow prediction by up to 11.25% while retaining foundational movement knowledge. Real-time robotic trials successfully guided actions such as picking up, moving, and placing a wok, supporting both stability and practical feasibility.
Overall, the work reframes EMG-driven assistance as a stable, transferable, and task-adaptive human–robot collaboration mechanism—not merely an intention-recognition problem. The researchers suggest future directions including more comfortable sensor designs, model lightweighting to reduce latency, and richer user feedback to accelerate adoption in home assistance and rehabilitation contexts.
Subject of Research: Unified three-level EMG control framework for elderly upper-limb robotic assistance (physiological, functional, behavioral levels).
Article Title: Artificial-Intelligence-Driven Electromyography Adaptation for Elderly Assistance at Physiological, Functional, and Behavioral Levels
News Publication Date: Jul 15, 2026
Web References: (Not provided)
References: (Not provided)
Image Credits: Jiaqi Xue, City University of Hong Kong.
Keywords: electromyography, EMG labeling, multi-joint intent decoding, knowledge distillation, assistive robotics, human–robot collaboration, collaborative robotic arm, elderly assistance
