Performance Comparison of the Nonlinear Bayesian Filters Supporting GPS/INS Integration

Y. Yi and D.A. Grejner-Brzezinska

Abstract:	The nonlinear Bayesian filtering techniques have been investigated for application in GPS/INS integration over the last few years as an alternative to the standard traditional Extended Kalman filter (EKF). The major objective of these efforts is to further improve the accuracy and the reliability of the currently implemented GPS/INS systems. This paper provides a preliminary performance analysis of two alternative nonlinear Bayesian filters, i.e. Particle filter (PF) and Sigma Point Kalman filter (SPKF), in comparison with EKF. The actual implementations of PF and SPKF are (1) PF with the EKF importance proposal distribution (EKF+PF for short), and (2) the Unscented Kalman filter (UKF), with a total of 24 states, including 9 navigation states (position, velocity and orientation) and 16 sensor error states (accelerometer and gyros biases and scale factor errors in 3 orthogonal axes) together with 3 level arm offset states. Kinematic experimental datasets from a high-end navigation grade INS (Honeywell H764G) and a low-cost consumer grade IMU (Crossbow MEMS IMU400C) were collected using a land vehicle. The datasets are used in this paper to provide the performance comparison of the above mentioned filters for both IMU sensors. The preliminary results shown here indicate that UKF is better than EKF in terms of the free inertial navigation performance for both the high-end and the low-end consumer grade inertial sensors. The advantages of UKF vs. EKF are more obvious in the low-end inertial sensor. However UKF is more computationally expensive than EKF, and the computational load increases linearly with the number of states in the state vector, in comparison with EKF. With a small number of particles, the free inertial navigation performance of EKF+PF is slightly better than that of UKF and EKF for the high-end inertial sensor. However, EKF+PF behaves worse than EKF and UKF for the lowend inertial sensor, which may indicate that the adapted number of particles in the EKF+PF implementation is not sufficient for this type of sensor.
Published in:	Proceedings of the 2006 National Technical Meeting of The Institute of Navigation January 18 - 20, 2006 Hyatt Regency Hotel Monterey, CA
Pages:	977 - 983
Cite this article:	Yi, Y., Grejner-Brzezinska, D.A., "Performance Comparison of the Nonlinear Bayesian Filters Supporting GPS/INS Integration," Proceedings of the 2006 National Technical Meeting of The Institute of Navigation, Monterey, CA, January 2006, pp. 977-983.
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In