Variational Bayesian Adaptive Embedded Cubature Kalman Filter Algorithm for Initial Alignment of SINS with Uncertain Observations

In the paper, a new variational bayesian based adaptive embedded cubature Kalman filter (VB-AECKF) for joint estimation of the dynamic state and measurement noise is proposed under a swing base. By constructing fully symmetric embedded integration rules for multidimensional integrals with a Gaussian weight function, a fifth-degree AECKF based filter with the estimated free parameter of ECKF based on maximum likelihood criterion is introduced in the initial alignment. The results of turntable experiment show that the proposed filter has better robustness to resist the uncertainties of measurement noise. What's more, the alignment accuracy and convergence rate of the VB- AECKF is much better than that of CKF.