Prediction of lower limb joint angle using sEMG based on GA-GRNN

Abstract

Gait analysis is an important research direction of the exoskeleton robot design. The ability to quickly and accurately calculate each joint angle of the lower extremities during level walking directly affects the accuracy of on-line control performance of the exoskeleton robot. As a direct biological signal from the brain, Surface Electromyography (sEMG) can reflect the motion intention of the human body ahead of muscle action. This paper presents a General Regression Neural Network tuned by Genetic Algorithm (GA-GRNN) based knee joint angle prediction model using sEMG signals. The proposed scheme can not only reduce errors due to manual parameter selection, but retains advantages of GRNN, e.g. excellent abilities of local approximation, nonlinear mapping, which greatly improves the prediction accuracy of the knee joint angle. To validate the algorithm, 5 subjects were selected to complete the level walking experiments to acquire sEMG and kinematics of lower limbs. After filtering and normalization, the signal was input into GA-GRNN model, the weight parameters were tuned by GA. The optimized model then can predict knee joint angle using sEMG signals. To improve the prediction accuracy, five-point smoothed cubic smoothing algorithm was used. The results show that the proposed GA-GRNN can achieve high estimation accuracy with less training time. It has great potential in lower limb exoskeleton robot field.