Abstract:The motion performance evaluation of human walking with an exoskeleton is an important guide for exoskeleton hardware iteration and control strategy optimization. Compared to the conventional experimental-based evaluation, that based on the dynamic simulation of the musculoskeletal-exoskeletal coupled system avoids the immense time and labor expense of platform establishing, experimenting and data processing. Moreover, further taking the human neuromuscular dynamics into account, the forward dynamic simulation of the coupled musculoskeletal-exoskeletal system can also generate physiological and biomechanical information such as muscle activation, muscle force, and joint torque that are unmeasurable during the experiment. This enables the muscle-level evaluation of human kinematics wearing the exoskeleton under various environments. Based on a comprehensive neuro-musculoskeletal-exoskeletal coupled dynamic simulation framework that integrates the human neuromusculoskeletal dynamics, foot-ground contact, and human-exoskeleton interaction, this study focuses on the human-exoskeleton coupled system and researched the kinematics of the system walking on the horizontal and uphill (5.71 ^\circ , 10%) treadmill at 0.8 m/s. The neuromuscular reflex parameters in the model are decided using the covariance matrix adaptation evolution strategy with the stepwise, multi-objective optimization, with multidimensional criteria considered, including the system motion, muscle expenditure, foot-ground reaction forces, etc., and then efficient system dynamic simulation can be accomplished without experiment data input. Results showed that the proposed framework can not only generate a full range of physiological and biomechanical signals including muscle activation, muscle force, joint torque, and joint angle, but also reflex and explain the adaptive change of muscle activation patterns to the walking environment variation. We further conducted experiments where the subject wore the exoskeleton walking under the same conditions, and compared the muscle activation with the measured sEMG signals. Results demonstrated that the muscle activities and their changing tendency concerning the ground slope in simulation and experiment maintained high consistency; The RMS values of activation of hamstrings, gluteus maximus, iliopsoas, and soleus when walking uphill increased significantly compared to those when walking horizontally, while that of rectus femoris and tibialis anterior reduced comparatively. The results above confirmed the effectiveness and accuracy of the proposed simulation framework in capturing qualitative variance of muscle activation patterns induced by terrain variation without experimental input. The study provides a new approach to conducting walking simulation and performance evaluation of the human-exoskeleton coupled system under multiple scenarios, and offers a theoretical reference and a simulation method for future research on hardware design and control strategy optimization of the exoskeleton.