Comparison of Isometric Force Estimation Methods for Upper Limb Elbow Joints

Abstract

For post-stroke patient, the interaction force between the impaired limb and the rehabilitation robotic system is an important issue for patient safety in clinical treatment. The surface electromyographic signals is generated from the center nervous system which can present the muscle activity for predicting the muscle force without force/torque sensors. For predicting the isometric joint force, there are two kinds of methods, the model-based approach and the model-free approach. In the model-based method, a musculoskeletal model would be usually used for calculating the individual muscle force. The interaction force would be calculated by a bio-mechanical model based on the individual muscle force. Conversely, the model-free would establish an approximate model by directly mapping the relationship between the force data and the sEMG signals. The purpose of this paper is to compare these two kinds of prediction method for interaction force between human and contacting environment by the sEMG signals. In this study, the isometric elbow joint extension motions are focused. Two muscles of the upper limb were selected to record EMG signals. The Hill-type musculoskeletal model was employed as the model-based approach and the neural network was selected as the model-free approach. The experimental results show that the model-free method have the better prediction performance which can provide acceptable prediction results with root-mean-square error (RMSE) error below 2.20 N.