Development and experimentation of an underwater reconfigurable vehicle for survey, inspection and intervention

The increasing demand for flexible and cost-effective solutions in underwater exploration and maintenance has driven the development of innovative marine robotic systems capable of adapting to a wide range of operational scenarios. This thesis presents the design, development, and experimental validation of the RUVIFIST (Reconfigurable Underwater Vehicle for Inspection, Free-floating Intervention, and Survey Tasks) vehicle, a novel Autonomous Underwater Reconfigurable Vehicle (AURV) capable of autonomously changing its shape to optimize performance across different mission objective. The RUVIFIST vehicle is developed to fill the gap between the traditional Remotely Operated Vehicles (ROVs), which offer high maneuverability but limited autonomy, and Autonomous Underwater Vehicles (AUVs), which excel in long-range endurance but lack precise control in in-site operations. The vehicle achieves this duality by exploiting a reconfigurable structure that allows it to operate both in a streamlined ``survey'' configuration and a compact "hovering" configuration, suitable for inspection and intervention tasks. The research encompasses multiple aspects of underwater robotics. The first part of the work presents the mechanical and software design of the RUVIFIST vehicle, focusing on modular interconnections, watertight enclosures, and a ROS-based architecture for distributed control and communication. The second part introduces the Guidance, Navigation, and Control (GNC) framework, including a Gain-Scheduled PID (GS-PID) controller to maintain performance consistency across different configurations. A Physics-Informed Neural Network (PINN) was developed to estimate hydrodynamic parameters such as added mass, and linear drag coefficients directly from on-field data, improving the fidelity of the dynamic model and the feed-forward control implemented along side the GS-PID. In parallel, an Autoencoder-based Fault Detection and Identification (FDI) system was implemented to detect and classify actuator or sensor faults in real time, increasing mission safety and robustness. Experimental validation was carried out both in controlled and open-water environments, confirming the vehicle’s capability to autonomously reconfigure its geometry and maintain stable navigation during transitions. Results aim to demonstrate accurate dynamic parameter estimation, enhanced control performance and also highlight the difference, in terms of added mass and linear drag, between "survey" and "hovering" configuration of the vehicle. In conclusion, the thesis contributes to the field of marine robotics by delivering one of the first fully operational prototypes of a reconfigurable underwater vehicle capable of performing both survey and intervention tasks thanks to the capability of autonomously change its shape. The combination of modular design, adaptive control, and AI-driven perception establishes a foundation for next-generation AURVs that can perform long-term resident operations, thus reducing deployment costs and extending the capabilities of autonomous subsea systems.