Development of a sEMG-Based Joint Torque Estimation Strategy Using Hill-Type Muscle Model and Neural Network

Purpose Exoskeleton control based on motion intention recognition has received increasing attention due to its promising prospect, and estimating joint torque is an effective approach to conduct intention recognition. In this paper, a surface electromyography (sEMG) signals-based joint torque estimation strategy is proposed to quantify the motion intention. Methods Different from the majority of existing torque estimation strategies, two major improvements have been achieved. System identification is presented to estimate elbow angle which can be used in the Hill-type muscle model, and hence, the use of angular transducer is replaced. Besides, neural network is used to train the optimal factor of muscle activation to make the estimated torque more accurate. Finally, static and dynamic experiments are conducted respectively to verify the effectiveness and improvements of this strategy in terms of torque estimation accuracy. Results Compared to the other two existing torque estimation strategies, results show that this method is proved to make some progress in respect of torque estimation accuracy under different experimental conditions. The correlation coefficient increases by 2–9%; root-mean-square error (RMSE) reduces by 0.2–2.5 Nm; normalized root-mean-square error (NRMSE) reduces by 0.5–9.5%. Conclusion The proposed torque estimation strategy could accurately identify the motion intention and reduce the use of angle sensor. Besides, it lays a foundation for rehabilitation exoskeleton robot control.