Calibration-free sEMG intention recognition via self-supervised pretraining and adversarial domain alignment for upper-limb rehabilitation

Comprehensive Summary

This study focused on improving calibration-free recognition of upper-limb movement intentions using surface electromyography (sEMG), a key challenge in rehabilitation engineering due to large inter-user and inter-session variability. The authors introduced a deep learning framework that integrates self-supervised pretraining and adversarial domain alignment to learn robust, transferable sEMG features without relying on labeled calibration data for each new user. Self-supervised learning enabled the model to capture meaningful time-frequency patterns of muscle activation from unlabeled signals, while adversarial alignment minimized differences across subjects and datasets. The framework was evaluated on established public datasets under leave-one-subject-out and cross-dataset conditions, closely mimicking real-world clinical use. The results showed consistent high classification accuracy and strong generalization to unseen users and recording setups, outperforming conventional calibration-dependent and calibration-free methods, with less signal noise and inter-subject variability.

Outcomes and Implications

These findings have significant implications for rehabilitation robotics, prosthetics, and assistive devices that rely on sEMG for control. Removing the need for repeated calibration sessions reduces clinician workload and shortens setup time, making these technologies more practical for routine clinical deployment and long-term home use. More accurate and reliable intention recognition can improve the responsiveness and intuitiveness of human–machine interaction, which is especially important for patients recovering from neurological injuries such as stroke, spinal cord injury, or neuromuscular disorders. By enabling plug-and-play sEMG systems that adapt across users, this approach supports more scalable, patient-friendly rehabilitation solutions and aligns with the growing push toward personalized, technology-assisted therapy outside traditional clinical environments.