EeShan Bhatt, Ph.D., 2021
Henrik Schmidt, Advisor
The global positioning system (GPS), ubiquitous for air- and terrestrial-based drones, cannot provide navigation for autonomous underwater vehicles (AUV). Yet, in ice-covered environments, underwater navigation is crucial for vehicle success. Current methods employ a deterministic sound speed to convert recorded travel time into range, ignoring how the vertical sound speed structure affects sound propagation and driving error into hundreds or thousands of meters.
This thesis demonstrates an embedded framework for a modular, real-time, physics-driven estimation of range from travel time. This enabled a successful AUV deployment in the Beaufort Sea, in March 2020, which faced the unique conditions of total ice-cover and a dynamic, complex acoustic propagation environment in the Beaufort Lens. By investigating the variability of the Beaufort Lens since 2006, designing a human-in-the-loop graphical decision-making paradigm to encode sound speed updates via a lightweight, acoustic message, and predicting a group velocity as a function of extrapolated source and receiver locations, we achieve a mean absolute ranging error of roughly 11 m, and further improve the group velocity prediction to rival GPS accuracy and surpass GPS precision. This work is the first to integrate a physics-based acoustic propagation model for real-time underwater navigation.