Visual–Acoustic-Based Framework for Online Inspection of Submerged Structures Using Autonomous Underwater Vehicles

Underwater inspections of critical maritime infrastructures are still predominantly performed by human divers, exposing them to safety risks and yielding limited accuracy and repeatability. Autonomous Underwater Vehicles (AUVs) offer a promising alternative by removing humans from hazardous environments and enabling systematic, repeatable inspection operations. However, current AUV systems lack the necessary autonomy and typically rely on prior knowledge of the environment, limiting their applicability in real-world scenarios. This study presents a visual–acoustic-based framework aimed at overcoming these limitations and moving a step closer to fully autonomous inspection operations using AUVs. Designed for cost-effective deployment on vehicles equipped with a minimal sensor suite—including a stereo camera, an acoustic range sensor, an Inertial Measurement Unit with magnetometers, a pressure sensor, and a Global Positioning System (used only on the surface)—the framework enables inspection of unknown underwater structures without human intervention. The main contribution lies in the integration of perception and navigation into a unified architecture, allowing the AUV to leverage the exteroceptive sensor not only for scene understanding but also to support real-time control and mission adaptation. Perception data are combined with proprioceptive observations to adapt motion based on the environment, enabling autonomous management of the inspection mission and navigation with respect to the target. Furthermore, a mission manager coordinates all phases of the operation, from initial approach to structure-relative navigation and visual data acquisition. The proposed solution was validated through a sea trial, during which an AUV autonomously inspected a harbor pier. The framework computed control actions in quasi-real-time to maintain a predefined safety distance, inspection velocity, and payload orientation orthogonal to the scene. These outputs were used online as feedback within the AUV's control loop. The underwater robot completed the inspection, maintaining mission references and ensuring effective target coverage, good-quality optical data, and consistent three-dimensional reconstruction. Overall, this experimental validation demonstrates the feasibility of the proposed framework and marks a significant milestone toward the deployment of fully autonomous AUVs for real-world underwater inspection missions, even in the absence of prior knowledge about the structure.