Point-Mass Filtering With Boundary Flow and Its Application to Terrain Referenced Navigation

Abstract

Point-mass filter (PMF) is a numerical Bayesian filtering algorithm that estimates the probability density of state variables using a deterministically defined grid on state space. An important factor that determines the performance of the PMF is how accurately one guesses the probabilistically significant region (called a nonnegligible region) where a grid will be placed. In this article, we introduce the concept of boundary to express the nonnegligible region and propose a method to define a grid tailored to the nonnegligible region of posterior density by transporting boundary according to log-homotopy induced flow. The proposed method ensures accurate estimation performance and robustness especially for uncertain prior information. Terrain referenced navigation (TRN) is composed of a nonlinear observation model defined by terrains and an uncertain initial position. Therefore, it is a major application of the PMF, which can handle nonlinearity and has a more robust characteristic than other numerical filters. In order to compare the proposed PMF with the conventional PMFs, this article applies the proposed method to TRN. Monte Carlo simulation results show that the proposed algorithm is more accurate and robust than the conventional PMFs.