Comparison of Forward and Reverse Decoding Methods of Knee Joint Moments based on sEMG

Abstract

It is important to calculate users' joint moment accurately for assistance efficiency assessment of exoskeleton robot. In this paper, to find out optimal model to calculate joint moment, the forward and reverse decoding model of knee joint moments (DMKJM) were established based on surface electromyography (sEMG), respectively. The forward DMKJM based on biological mechanism was established using Hill muscle model. The reverse DMKJM is based on Radial Basis Function (RBF) network. The performance of two models were compared at different motions data from human gait experiment. The results showed that the mean of Pearson correlation coefficient (p) of the reverse DMKJM based on RBF network (DMKJM-R) and the forward DMKJM based on Hill model (DMKJM-H) were 0.9345 and 0.8698, respectively. And the mean standard deviation (SD) of $p$ of DMKJM-R and DMKJM-H were 0.0239 and 0.0599, respectively. The mean and the mean SD of normalized root mean square error (NRMSE) of DMKJM-R were 0.0880 and 0.0157 respectively, while DMKJM-H were 0.1279 and 0.0269. Besides, the complexity of DMKJM-R was also less than DMKJM-H. In conclusion, the comprehensive performance of DMKJM-R is better than DMKJM-H, which can calculate joint moments more accurate and improve accuracy of assistance efficiency assessment for exoskeleton robots.