A novel Cubature Kalman filter based on switching Gaussian and student's t distribution
Zhecheng Zheng, Mingyuan Zhai, Yanhui Tong, Lidong He · Journal of Control and Decision · 2025
This paper proposes an adaptive robust nonlinear filter, named Gaussian-Student's t switching Cubature Kalman Filter (GSCKF), where the noise model switches between Student's t and Gaussian distribution. A Chi-squared test method is used to determine the noise distribution function according to the innovation. The proposed Chi-squared test dynamically adjusts the proportional coefficients of update weights based on the test values, which prevents the complete acceptance or rejection of observations at any given time. The Kullback-Leibler Divergence (KLD) fitting and the moment matching techniques are employed to achieve a smooth transition between Gaussian and Student's t-distributions to guarantee the filter continues to operate effectively after switching between noise modeling approaches. Finally, by comparing the proposed filter with other filters, it is demonstrated that our filter performs better in terms of high accuracy and lower computation burden.