A Three-Step Hill Neuromusculoskeletal Model Parameter Identification Method Based on Exoskeleton Robot

Different from functional alternative equipments such as prostheses, the highly human-machine collaboration performance are required in exoskeleton robots. At present, the commonly used physical human-machine interface can only obtain motion after the movement occurs. Therefore, it is difficult to predict the non-rhythmic movement. The inherent intent detection hysteresis exist in the physical human-machine interface. The cognition based human-machine interface directly detect the neural electrical signals of the human body, and has the characteristics of advanced motion detection. However, the existing methods based on the Hill muscle mechanics model mostly set muscle parameters based on experience. It is difficult to accurately predict joint torque of different subjects because of the parameter difference of the muscle model. Therefore, in this paper, we proposes a three-step muscle parameters identification paradigm for Hill muscle model based on exoskeleton robots. Then, the Adam optimizer with variable learning rate is employed to identify the muscle parameters. Eight healthy subjects are participated in the experiment. The results show that the proposed Adam optimizer with variable learning rate can make the parameters stably convergence. The estimated torque of the identified Hill muscle model have lower error than that of the neural network-based method. The performance of the proposed method is competitive with that of the State-of-art method.