Decoding Wrist and Hand Movement Using a Synergy-Driven, Variable Stiffness Musculoskeletal Model Under Different Loads

Myoelectric interfaces, which translate electromyography (EMG) signals into motor intentions, are extensively utilized in prosthetic control due to their intuitive nature. However, current methodologies extremely suffer from performance degradation under non-stationary EMG signals. Varying loads significantly alter muscle activation patterns, with variations in loading weights being a significant factor affecting performance. Conventional EMG-driven musculoskeletal models (MMs) demonstrate robustness across different loads but often neglect the influence of changes in muscle co-contraction levels induced by increased loads. To address this issue, we proposed an improved MM that integrated variable muscle and joint stiffnesses with synergy-based muscle activations into the muscle contraction dynamics. This model adapted to alterations in muscle co-contraction levels under diverse loads, allowing for robust estimation of hand and wrist movements without requiring additional training for different loads. Both offline and online experiments were conducted. The outcomes demonstrated that the proposed model significantly improved robustness and estimation accuracy compared to traditional envelopedriven and constant stiffness MMs. It maintained stable performance across different loads with training conducted solely under zero-load conditions, thereby reducing the training burden. This study presents a novel approach to enhance the reliability of EMG-based human-machine interfaces.