Lower Limb Joints Torques Continuous Estimate Model Based on Muscle Synergy for Patients With Motor Dysfunction

Accurate and reliable estimation of joint torques in the lower limb is essential for assisted control of rehabilitation exoskeleton robots that help patients recover. Surface electromyography is a valuable tool for revealing human motion. According to the myoelectric signal generation mechanism, muscle mechanical process, and muscle synergy theory, this study developed a model at the neural control level that relates myoelectric signals and lower limb joint torques. First, a muscle bioelectrical activation model based on electromyography signals is established. Second, a motor nerve activation model based on muscle synergy was developed to characterize muscle activation as muscle synergy structure and muscle synergy activation coefficients. Then, a continuous estimation model for lower limb joint torques with multi-scale feature fusion enhancement is established by integrating deep learning methods with the self-attention mechanism. The adversarial transfer learning method is used to optimize the model, enabling it to adapt to long-term use by patients during rehabilitation movements. Experimental results involving eight patients with lower limb motor dysfunction demonstrate that our proposed model can accurately estimate the torques of the hip and knee joints. It maintains strong performance over long-term use, with decision coefficients of 0.92 ± 0.06 for hip joints and 0.95 ± 0.03 for knee joints. This study lays the groundwork for further development of on-demand assisted control for exoskeletons.