Deterministic and stochastic Bayesian methods in terrain navigation

Terrain navigation is an application where inference between conceptually different sensors is performed recursively online. In this work the Bayesian framework of statistical inference is applied to this recursive estimation problem. Three algorithms for approximative Bayesian estimation are evaluated in simulations, one deterministic algorithm and two stochastic. The deterministic method solve the Bayesian inference problem by numerical integration while the stochastic methods simulate several candidate solutions and evaluates the integral by averaging between these candidates. Simulations show that all three algorithms are efficient and approximately reach the Cramer-Rao bound. However, the stochastic methods are sensitive to outliers and the deterministic method has the limitation of being hard to implement in higher dimensions.