• Reconfigurable vehicles
• Underwater robotics

In the last decades, the demanded tasks for subsea operations have become more and more challenging, for instance, intervention, maintenance, and repair of seabed installations, in addition to surveys. Currently, these tasks are usually carried out in hostile environments using Remotely Operated Vehicles (ROVs) deployed from a large surface vessel; however, ROV operations require high expenses due to the daily ship cost and the large number of crew members that such operations need. As a result, in order to reduce the inspection cost and deal with several logistic constraints, the interest in Autonomous Underwater Vehicles (AUVs) has rapidly increased in the last few years, becoming fundamental tools for marine scientists and industries to explore and monitor underwater areas. More in particular, AUVs are currently exploited for long-distance monitoring and inspection, as they are provided with an ever-increasing precision navigation system. As a drawback, AUVs are commonly not provided with high manoeuvrability in all the Degrees Of Freedom (DOFs). Therefore, the development of vehicles that can incorporate both the AUV inspection and ROV intervention functionalities can be considered a challenging task in the underwater industry as well as the scientific community. For instance, Autonomous Underwater Reconfigurable Vehicles (AURVs) can select the most appropriate configuration so as to dive with the optimal fluid dynamic efficiency and achieve the minimum power consumption for long-distance navigation (namely, in AUV modality); furthermore, the robot may be provided with a larger number of degrees of freedom for complex manipulation operations or to perform autonomous docking in a subsea station as an ROV. Therefore, AURVs do arise as a promising tool for incorporating several reconfigurable modules and accomplishing autonomous inspection, surveys, and intervention tasks. Motivated by the previous works, the RUVIFIST (Reconfigurable Underwater Vehicle for Inspection, Free-floating Intervention and Survey Tasks) AURV, capable of efficiently reconfiguring its shape according to the task at hand, has been designed and developed during the Ph.D. research work. More in detail, this vehicle has been provided with the capability to actively switch between a “hovering/opened” configuration, with an isotropic behaviour proper to face intervention operations, and a “survey/closed” one, with an elongated body that minimises the hydrodynamic damping force in the longitudinal direction and, as a consequence, reduces the power consumption for long mission tasks. The solutions proposed in this thesis have been firstly validated with realistic simulations made in order to check the different dynamic properties of the several vehicle configurations. Subsequently, preliminary dry and wet tests have been performed in the Mechatronics and Dynamic Modeling Laboratory (MDM Lab) of the Department of Industrial Engineering of the University of Florence (UNIFI DIEF), Italy. Several experimental campaigns have also been conducted to test the developed Guidance, Navigation, and Control (GNC) subsystems. Meanwhile, Industry 4.0 has been arising as a novel, cutting-edge era to be renowned as the fourth revolution: a revolutionary, innovative age where Automation, Artificial Intelligence (AI), as well as Robotics merge their respective scientific. As a result, an innovative strategy, based on Deep Learning (DL) approaches, was investigated, developed and implemented to estimate the vehicle model and employ it in the navigation algorithm. More in particular, the usage of Neural Networks (NNs) in underwater robotics is mostly related to solving Automatic Target Recognition (ATR) tasks; conversely, we investigate whether it is possible to employ NNs to estimate a dynamic model of the vehicle and use it in the navigation strategy. A preliminary offline validation is proposed in this thesis by using navigation datasets provided by the National Oceanography Center (NOC). Motivated by the obtained results, the usage of DL for AURV will be investigated in the near future. As a summarising statement, the designed and developed RUVIFIST AURV has emerged as a possible solution to merge the AUV capabilities of performing long-range autonomous navigation, based on efficient power-consuming torpedo shapes, with the ROV-related enhanced mobility requested in closer inspection and/or intervention tasks. Indeed, employing such AURVs can unavoidably change the current underwater robotic paradigm, based on a marked decoupling between AUV and ROV operations, toward unified and cutting-edge solutions. Nonetheless, there is still room for improvement: for instance, the usage of model-based controllers may be the key to providing the robot with a more performing control; furthermore, depending on the available onboard payload, state-of-the-art DL-based target recognition strategies can be implemented in the context of a task-oriented morphing capability of the vehicle. These issues, whose addressing constitutes a natural continuation of the research activity carried out so far, will be subjected to further investigation.