Aiming at the problems of insufficient real-time human-computer interaction and limited fatigue monitoring of lower limb rehabilitation exoskeleton, a multi-modal evaluation of lower limb muscle fatigue and exoskeleton variable stiffness impedance control method are proposed. This method extracts the characteristics of surface electromyography time-frequency domain and A-mode ultrasound muscle thickness, combines ventilation and subjective fatigue scale, and uses support vector machine to construct a five-classification model of fatigue level to achieve state evaluation. In addition, this study further designs a variable stiffness impedance control strategy to dynamically adjust the exoskeleton parameters in real time according to the muscle state to improve the interaction safety. Experiments show that the classification accuracy of the multi-modal fusion model proposed in this study is 92.30%, and the system delay is less than 200 ms. In the walking test, the variable stiffness control strategy can further significantly reduce the net respiratory metabolic rate of the user compared with no exoskeleton or under built-in control assistance of exoskeleton. The method proposed in this paper realizes effective fatigue assessment and adaptive assistance, and provides a new idea for intelligent interaction of lower limb rehabilitation robots.
