Terrain-Aided Navigation Based on Sequential Multiple Distribution Estimation Filter: Theory and Experiment

Particle filter (PF) is an effective way to deal with terrain-aided navigation (TAN) problems of autonomous underwater vehicles (AUVs), but it is susceptible to the effects of particle degeneracy and impoverishment. Although the multiple distribution estimation filter (MDEF) is robust to particle degeneracy and impoverishment, its filtering performance may be affected by the uncertainty introduced by distribution selection and the linearization error caused by the calculation of the probability density functions (PDFs) using the extended Kalman filter (EKF). To improve the filtering performance of the MDEF, this article proposes a sequential multiple distribution estimation filter (SMDEF) for TAN of AUVs. The SMDEF updates the weight via a recursive form, which allows distribution selection to be performed only when particle degeneracy occurs, thus mitigating the uncertainty introduced by distribution selection. Meanwhile, the SMDEF uses the intermediate distributions to calculate the likelihood and posterior PDFs under progressive Bayesian filtering framework, thus reducing the linearization error of the EKF. Field trials demonstrate that the proposed SMDEF-based TAN can improve the average positioning accuracy by 14.9%, 12.9%, 45.6%, 42.7%, 25.5%, 48.9%, and 71.8%, respectively, compared with the TANs-based MDEF, progressive extended Kalman PF, cubature PF, robust fusion PF, robust PF, intelligent PF, and PF, which verifies the superiority of the proposed method.