An Improved MCMC-Based Particle Filter for GPS-Aided SINS In-Motion Initial Alignment

Abstract

In many practical applications, the initial alignment of the strapdown inertial navigation system (SINS) must be performed under large misalignment angles. However, the error model of SINS in this situation is nonlinear, making it hard to obtain an accurate initial alignment result. Aiming at solving this problem via nonlinear filtering, an improved Markov chain Monte Carlo (MCMC)-based particle filter (PF) algorithm is designed in this article, which uses a square-root cubature Kalman filter (SCKF), PF, and MCMC algorithm synthetically. The algorithm is dedicated to extract particles from the posterior distribution of the state by the improved MCMC algorithm and avoid the particle degeneration and the particle impoverishment in a PF. This can effectively improve the performance of particle filtering for nonlinear systems. Furthermore, to solve the problem caused by the fixed number of iterations in the typical MCMC algorithm, a novel adaptive approach is introduced according to the maximum mean discrepancy. The simulation and experimental results of GPS-aided SINS in-motion initial alignment show that the proposed algorithm can provide good filtering accuracy and have good real-time performance.