Researchers at the City University of Hong Kong and the Beijing Institute of Technology have developed a 3-level AI-driven electromyography (EMG) control framework that translates muscle signals into robotic assistive actions for older adults. Published in Cyborg and Bionic Systems, the study details how the system bridges physiological signal labeling, functional multijoint intent decoding, and behavioral task adaptation to help seniors maintain independence in daily activities like cooking and carrying groceries.
Overcoming EMG Control Challenges in Real-World Scenarios
As people age, declines in muscle strength, coordination, and sensory function make daily tasks like doing laundry, getting dressed, cooking, and using tools increasingly difficult. According to City University of Hong Kong researcher Jiaqi Xue, bringing EMG-based control into real-time daily use has historically faced major roadblocks. Traditional setups rely on complex manual annotation or external motion-capture systems to map signals to motor intent. Furthermore, most existing models focus on single-joint movements rather than coordinated multijoint actions involving both the hand and the elbow.
EMG signals also vary significantly across different tasks, postures, and application environments. This variability often causes models to struggle with new tasks while forgetting previously learned capabilities. To solve these issues, the research team designed a unified framework that addresses signal labeling, intent decoding, and task adaptation simultaneously, according to findings published in the journal.
Did you know? Electromyography (EMG) signals reflect neuromuscular control and typically appear before visible limb movement occurs, making them a natural interface for controlling assistive robots.
A Three-Level Approach to Human-Robot Collaboration
The newly proposed framework operates across three distinct levels: physiological, functional, and behavioral. At the physiological level, the researchers gathered upper-limb EMG data from younger and older participants, focusing on four specific muscles: the triceps, biceps, extensor digitorum, and flexor digitorum superficialis. To cut down on reliance on manual annotation and external capture gear, they deployed selective active labeling and contextual labeling strategies based on muscle activation traits.
According to the study, selective active labeling works best for hand opening and closing, where EMG signals show short and strong fluctuations. Contextual labeling suits elbow flexion and extension, where muscle activation remains more continuous. At the functional level, the team built a one-dimensional convolutional neural network. This network simultaneously predicts hand states (relaxed, open, closed) and elbow states (relaxed, flexed, extended) from 4-channel EMG signals, utilizing a voting mechanism to stabilize real-time outputs.
Achieving High Accuracy in Multijoint Intent Decoding
Performance tests revealed strong results across the board. At the functional level, the one-dimensional convolutional neural network achieved offline accuracies of 95.42% for hand state prediction and 93.97% for elbow state prediction, according to the published data. During real-time multijoint coordination tests covering nine hand-elbow motion combinations, the model reached an overall accuracy of 95.34% and successfully drove smooth robotic assistance.
At the behavioral level, the researchers introduced knowledge distillation. This technique allowed the model to master new daily tasks—such as cooking, lifting bags, and pulling grocery carts—while retaining basic joint-movement knowledge. In testing, the student model improved hand and elbow prediction by up to 11.25% during cooking tasks compared to the teacher model.
Pro Tip: When designing assistive robotics for elderly care, integrating knowledge distillation prevents catastrophic forgetting, allowing the AI to acquire new domestic skills without losing foundational motor controls.
Real-Time Robotic Validation and Future Outlook
To validate real-world feasibility, the research team deployed the framework in a live robotic control system combining EMG acquisition, edge computing, and a 6-axis collaborative robotic arm. Experiments demonstrated that the system successfully interpreted user intentions to pick up, move, and put down a cooking wok.
Authors of the paper include Jiaqi Xue, Ziqi Li, Xiaoyang Zou, Zijia Qu, Shengjie Yang, Colin Pak Yu Chan, Yanchen Liu, Zhou Zhao, Jing Zhang, Clio Yuen Man Cheng, Haiyang Wang, Kehan Zou, Yafei Zhao, Vivian Weiqun Lou, Ning Xi, and King Wai Chiu Lai. The work received partial funding from the Research Grant Council of the Hong Kong Special Administrative Region Government under TBRS Grant T42-717/20-R and CRF Grant C7100-22G.
“In the future, we will combine flexible sensors to enhance wearing comfort, optimize model lightweighting and system latency, and introduce more user feedback and interaction mechanisms to promote the practical application of such systems in home assistance, rehabilitation training, and healthy aging scenarios,” Jiaqi Xue stated.
Frequently Asked Questions
What is a 3-level AI-driven EMG control framework?
It is an assistive robotics architecture that processes muscle signals across physiological labeling, functional multijoint intent decoding, and behavioral task adaptation to control robotic arms for elderly users.
Which muscles are monitored in this EMG control system?
The system monitors four key upper-limb muscles: the triceps, biceps, extensor digitorum, and flexor digitorum superficialis.
How does knowledge distillation help assistive robots?
Knowledge distillation enables the AI model to learn new complex daily tasks—such as cooking or carrying groceries—without forgetting previously acquired basic joint-movement knowledge.
What funding supported this research?
The study was supported by grants from the Research Grant Council of the Hong Kong Special Administrative Region Government, including TBRS Grant T42-717/20-R and CRF Grant C7100-22G.
What are your thoughts on using EMG-driven exoskeletons and collaborative robotic arms for elderly care? Share your opinions in the comments below, explore our related articles on bionic engineering, or subscribe to our newsletter for the latest updates in assistive technology.