Robust cubature Kalman filter based on maximum correntropy and resampling-free sigma-point update framework

Abstract

The problem of nonlinear filtering for GNSS/INS under non-Gaussian noise is studied in this paper, where an improved maximum correntropy cubature Kalman filter (MCCKF) is proposed by utilizing resampling-free sigma-point update framework (SUF). The MCCKF has better robustness than CKF in case non-Gaussian noise appears. However, only the Gaussian moments are circulated in MCCKF, which is less efficient when the nonlinearity of dynamic function is strong. The proposed algorithm named RMCCKF, imports the re-scaled measurement and state prediction covariance that produced by maximum correntropy criterion into the direct construction of posterior sigma points. By constraining the posterior covariance of MCCKF and modifying the sigma points directly, the resampling-free SUF spreads extra information of dynamic model over the posterior sigma points update. The proposed algorithm is verified by using numerical simulation and field test on GNSS/INS. Experiment results indicate that, RMCCKF achieves better performance than CKF and MCCKF, especially when the filtering problem is of high dimension. On the GNSS/INS of land vehicle navigation, RMCCKF not only improves the position but also reduces the heading error from 1.77 degree to 0.26 degree, compared with MCCKF.