An improved regularized particle filter for GPS/INS integration

Hybridization techniques receive a renewed interest due to recent navigation systems such as Galileo. Hybridization takes advantage of the complementarity of different sensor types to increase navigation performance. This study focuses on the integration of the Global Positioning System (GPS) and the inertial navigation systems (INS). GPS allows to compensate for the long term drift of INS estimates, while aided INS provide a solution in case of GPS signal blocking. The GPS/INS coupling is performed by a nonlinear filtering approach whereby GPS measurements are used to correct INS estimates. However, a classical particle filter is bound to diverge due to the dynamics of the unknown parameters. Indeed, the state model has a small process noise and is exponentially unstable. The regularized particle filter presented in N. Oudjane et al. (2000) allows to overcome this limitation at the expense of an increased estimation variance. This study proposes to introduce a Metropolis-Hastings step to accept/reject the particles updated by the regularization process. The method is shown to prevent the degeneracy without introducing additional noise on the estimates.