Undersea Active Terrain-Aided Navigation (ATAN)

A significant limitation for Terrain Aided-Navigation requires an accurate a priori terrain map. We introduce a solution called Active Terrain-Aided Navigation that utilizes an optimal, information-theoretic architecture in conjunction with reinforcement learning to solve a stochastic dual exploration-exploitation optimization problem. The dynamic path planning solution balances competing navigational objectives of total area coverage and position localization. The exploration component uses Multiple Gaussian Process Models while the exploitation component uses a combination of a novel entropy-based approach for information discovery and prioritization coupled with a maximum a posteriori estimation technique. Simulation results and analysis for an autonomous underwater vehicle are presented.