Titolo della tesi: Responsive Navigation, Realistic Trials: Adaptive Autonomous Underwater Vehicles Acoustic Positioning Techniques and Communication-Aware Cyber-Physical Simulation.
The oceans remain the least explored biome on Earth, as the robotic and networking technologies that have empowered land, air, and space completely loose their effectiveness, since radio waves do not penetrate water.
As of today, the already constrained acoustic channel is the only mature technology that can sustain the two main challenges of underwater systems: localization and communication.
Moreover, testing novel approaches at sea involves costly operations and equipment, and the noisy and time-varying nature of the marine environment severely hinders the ability of performing controlled and rigorous experimental campaigns.
This thesis addresses two closely related obstacles to autonomous underwater exploration: (i) the need for autonomous, accurate localization in non-stationary environmental conditions, and (ii) the absence of realistic, low-cost platforms to test and refine such capabilities before committing to expensive and unpredictable sea trials.
An analytically grounded acoustic positioning framework for Autonomous Underwater Vehicles (AUVs), validated in real deployments, is first proposed, exposing how even simple, deterministic models can yield meter-level accuracy.
In depth analyses, however, reveal how static noise modeling undermines reliability under time-varying environmental and sensors noise.
To address this, this thesis introduces an adaptive localization architecture that is able to tune both process and sensor models, corrects for varying noise conditions, and harmonizes asynchronous acoustic ranges with inertial and complementary data using probabilistic filtering techniques.
This approach achieves robust performance, improving localization accuracy to more than 50%$with respect to deterministic solutions, without requiring laborious calibration, even in the presence of non-stationary disturbances.
Yet, evaluating complex solutions that aim at precisely characterizing the noise parameters of the system demands testbeds that are able to capture the environmental nuances they mimic.
Field trials, though invaluable, are costly and often unsuitable for repeatable, large-scale algorithm development.
To fill this gap, this thesis proposes MAREA, the first cyber-physical co-simulation framework bridging high-fidelity vehicle dynamics, and realistic underwater communications and sound propagation.
By tightly coupling a physics-aware robotic simulator with a modified ns-3 engine that supports multimodal underwater links and stack reconfiguration, MAREA provides a realistic "digital ocean" for experimentation.
Its physically grounded modeling of hydrodynamics, noise, and synchronization effects bridges the gap between in silico accuracy and field uncertainty.
Together, these contributions advance the foundations of underwater autonomy, offering both a framework able to model the nuances typical of real scenarios and methods that adapt to reality.