The development of underwater high-resolution mapping plays a key role in scientific and technical areas as geology and marine biology, underwater archeology or inspection of energy infrastructure facilities such as oil fields or offshore wind turbine areas. The state of the art of underwater mapping systems based on AUVs, allows the development of 2D maps (e.g., side scan sonar and optical or acoustic photomosaic) and 2.5D (e.g., bathymetry sonar). Although the progressive incorporation of optical sensors (i.e., laser scanners, stereoscopic cameras, monocular cameras and structure from motion techniques) opens the door to the development of 3D maps, the simplicity of the trajectories executed by current robots, that only follow the profile of the seabed at a certain altitude, makes it imposible to safely explore areas with a high 3D relief. Thus, the automatic production of 3D maps not only requires the ability to econstruct the 3D geometry of the environment from gathered data but the necessary skills required by the robots to explore in a safe, efficient and comprehensive way the geometry to collect this data. If we also consider that the DVL (Doppler Velocity Log), the main localization sensor of an AUV, does not work correctly in slopes greater than 30 ° (i.e., in the areas with high relief that we want to map), we face an interesting technical/scientific challenge beyond the current state of the art. Therefore, the 3DAUV project proposes the development of techniques to move beyond the state of the art, developing a methodology to explore underwater areas, a priori unknown, and with a strong 3D relief. The AUV Girona 500 will be equipped with a new payload that incorporates a scanner sonar, capable of providing 3D
point clouds and an omnidirectional camera that will provide a hemispheric vision. A view planner will be used to propose the next-best viewpoint where the AUV has to move in order to capture a new 3D point cloud to discover the 3D geometry of the explored region. A SLAM module will be used to estimate the graph of positions that compose the path traveled by the robot. This graph will be build from the AUV navigation data, updates from a USBL sensor, and 3D point cloud registers with sufficient overlap. Point clouds will be merged into a volumetric model in the form of a three-dimensional occupancy grid over which a second planner, in this case a coverage path planner, will generate trajectories to ensure a thorough exploration of the 3D model surface taking into account the camera field of view. Finally, a task planner will opportunistically intertwining the paths proposed by both planners to optimize the exploration time, the energy and the localization accuracy. The proposed system will be experimentally demonstrated by performing the inspection of two environments with significant relief: a seamount and an underwater cave.
