Indoor spatial navigation is a challenge for individuals with limited visual input (Low Vision) and can be a challenge for individuals with normal visual input in unfamiliar, complex environments. Although outdoor navigation can be achieved using Global Positioning Systems (GPS), these signals are not available for indoor navigation. We have developed an indoor navigation aid that uses a Bayesian approach (Partially Observable Markov Decision Process; POMDP) to localize and guide a user within an unfamiliar building. We describe the fundamental concept here, in addition to empirical evaluation of the system within virtual environments of varying sizes. Our findings show that the system improves performance when visual information is degraded and also improves performance under normal conditions in large, complex environments.