Prediction of joint angle by combining multiple linear regression with autoregressive (AR) model and Kalman filter

Abstract

In this paper, a new prediction algorithm combining multiple linear regression with autoregressive model and Kalman filter (MLRAR-KF) is proposed to predict the elbow joint angle. The MLRAR model updating weights with Kalman filter is shown to be able to predict joint motion with high accuracy and well robustness. In comparison to existing prediction algorithms, MLRAR-KF can predict joint angle with higher accuracy and better robustness. A data acquisition system was used to collect sEMG and elbow joint angle signals of human upper limb. The experimental results demonstrate the benefits of MLRAR-KF prediction algorithm. Comparison of computational complexity about some existing prediction methods and MLRAR-KF is conducted to analyze the real-time performance.