Bionic soft robotic glove with EMG-based gesture and grip strength synchronized prediction for grasping assistance

Wearable hand-assistive robotics play an important role in aiding elderly patients with hand dysfunction, where accurate gesture recognition and grip strength estimation are essential for natural human–robot interaction. However, few studies have tackled both tasks simultaneously. Inspired by the biological tendon-muscle system, this work introduces a soft robotic glove actuated by tendon-sheath artificial muscles. The system features an EMG-based controller that provides real-time assistance by jointly predicting hand gestures and grip strength using a GRU-based domain-adversarial neural network with a composite loss function, enabling combined classification and regression from shared EMG features. The model achieved 92.12% gesture classification accuracy and an R² of 0.935 within subjects, and 79.43% accuracy with an R² of 0.80 across subjects. Real-time testing with an unseen user further confirmed the model's robustness, achieving 80.94% accuracy and an R² of 0.86. The soft robotic glove also significantly reduced forearm flexor muscle activity by up to 46.9% during grasping tasks, demonstrating effective assistance. Overall, this EMG-driven soft robotic glove offers personalized, adaptive, and precise hand support, showing strong potential to enhance autonomy and quality of life for elderly users.