Application of Unscented Kalman Filter with Non-symmetric Sigma Point Sampling on the Integrated Navigation System

The integrations of a global positioning system (GPS) and an inertial navigation system (INS) usually use error state models with linear or non-linear Kalman filters. In high dynamic environments, these systems introduce errors in the system models due to linearization. To overcome this drawback, the unscented Kalman filter (UKF) is applied on the non-linear total state model of the GPS/INS integrated system because the UKF is a nonlinear estimator that is particularly well suited for complex nonlinear systems. The sigma points in UKF are usually sampled symmetrically around the mean value and the random variable to be transformed is assumed to be Gaussian. If the variables depart from Gaussian nature, the performance of the system is degraded. To enhance the UKF performance the non-symmetric sigma points sampling is addressed in this paper. The non-symmetric factors for sigma points are carefully chosen to avoid local and global sampling problems. The simulations are done using real INS and GPS data for symmetric UKF (SUKF) and non-symmetric UKF (NSUKF) algorithms. The navigation performance and robustness of the proposed algorithm are also compared with that of the SUKF. According to the simulation results from the application of NSUKF on a nonlinear total state model, the performance and robustness of the navigation system is significantly improved under the environment with a number of satellites less than four. Hence, NSUKF is better choice for low cost INS/GPS integrated navigation systems and a good alternative for SUKF