Adaptive Approximate Bayesian Computational Particle Filters for Underwater Terrain Aided Navigation

To perform long-term and long-range missions, underwater vehicles need reliable navigation algorithms. This paper considers multi-beam Terrain Aided Navigation which can provide a drift-free navigation tool. This leads to an estimation problem with implicit observation equation and unknown likelihood. Indeed, the measurement sensor is considered to be a numerical black box model that introduces some unknown stochastic noise. We introduce a measurement updating procedure based on an adaptive kernel derived from Approximate Bayesian Computational filters. The proposed method is based on two well-known particle filters: Regularized Particle Filter and Rao-Blackwellized Particle Filter. Numerical results are presented and the robustness is demonstrated with respect to the original filters, yielding to twice as less non-convergence cases. The proposed method increases the robustness of particle-like filters while remaining computationally efficient.