A model-free cubature Kalman filter(MF-CKF) combined with Gaussian process regression(GPR) is presented.
Cubature Kalman filter(CKF) is a new nonlinear Gaussian filter, which is superior than uncented Kalman filter(UKF).
Gaussian process regression is introduced into cubature Kalman filter to overcome precision decreasing caused by model
uncertainty. Gaussian process is applied to establish nonlinear models by using training data, which efficiently avoids the
degradation of filtering performance. Simulation results show the superiority of MF-CKF in the case of model uncertainty.