Adaptive hybrid Kalman filter based on the degree of observability

Kalman filter is generally selected as the data fusion algorithm in the integrated navigation system of Autonomous Underwater Vehicles (AUVs). The output correction method does not correct the system mathematical model so that navigation errors are gradually accumulated. Frequently performing feedback correction of full states will reduce the convergence and even cause divergence. Therefore, the hybrid correction method is usually applied in the practical system by combining the output correction method with the feedback correction method. However, the divergence still occurs in the incompletely observable system. This paper presents a new adaptive hybrid Kalman filter based on the degree of observability analysis of system states. The degrees of observability are defined from the viewpoint of error attenuation of the initial state, which are normalized and defined as feedback factors. Feedback factors adaptively modify feedback values of state estimations in the hybrid Kalman filter. The proposed filter is applied in the attitude determination based on IMU, and the test results indicate that the new method can effectively inhibit divergence and improve the accuracy of the incomplete observable system.