Differentiable Dynamic Model for Musculoskeletal Simulation and Exoskeleton Control

The exoskeleton, a wearable device, was designed based on the physical and cognitive interactions for the users. The control of exoskeleton was based on biomedical signal from the user's intention, and outputs calculated from the algorithm to make the movement smoothly. However, transmission from input of biomedical signal, such as electromyography (EMG) to application on the torque or angle of the exoskeleton has limitation of time-lagging or precisely predicted trajectory, which make the mismatch between the subjects and exoskeleton. In this study, we proposed an EMG-based single-joint exoskeleton system, merged a differentiable continuous system with a dynamic musculoskeletal model. The parameters of each muscle contraction were calculated, and further applied on the rigid exoskeleton system to predict precise trajectory. The results revealed the well torque and angle prediction of the knee exoskeleton, and good performance of assistance during movement. Furthermore, the convergence and execution time in our purposed method had better performance than the other models. Conclusively, a differentiable continuous system merged with a dynamic musculoskeletal model provided an effective and accurate performance for exoskeleton controlled by the EMG signals.