INS/GPS Integration Using Gaussian Sum Particle Filter

In INS/GPS integration, the data fusion algorithm involves properly handling of nonlinear models. Therefore the nonlinear filtering methods have been commonly applied in the INS/GPS integration to estimate the state vector. The most popular and commonly used method is the Extended Kalman Filter (EKF) which approximates the nonlinear state and measurement equations using the first order Taylor series expansion. On the other hand, recently other nonlinear filters such as Particle Filter (PF), Unscented Kalman Filter (UKF) and Gaussian Sum Filter (GSF) are also considered for use in the INS/GPS integration. The GSF is a nonlinear filter where its predictive priori density is assumed to be the sum of several normal distributions. However the first order Taylor series approximation is applied in the GSF for updating each distribution similarly to the EKF. So there exists the possibility of degrading the filtering performance under high nonlinearity shown in the classic EKF. In this paper we apply a nonlinear filter combining the GSF and PF, which is referred to as Gaussian Sum Particle Filter (GSPF). The GSPF is based on the similar concept of the GSF, but the GSPF updates its Gaussian sum expressions by using the particles instead of the linear approximations. The performance of the GSPF based loosely coupled INS/GPS integration is compared with other filters in numerical simulations. From the simulation results, it is found that the GSPF has an ability to improve the navigation performance when there exist large uncertainties in the initial estimates.