Towards an Autonomous, Visual Inspection-aware 3D Exploration and Mapping System for Water Ballast Tanks of Marine Vessels

While there exist several approaches for autonomous exploration of confined spaces, when it comes to inspection tasks, there are not many works that take into account the actual classification of defects in the robot’s view. Such approaches may result in less than optimal observations –even miss corrosion areas– and, therefore, can hinder the fulfillment of the purpose of the exploration, which is the identification of defects. In this paper, the first steps towards the unification of the exploration and inspection procedures, are theorized and experimentally evaluated. The next best view algorithm is augmented with information stemming from the defect classification. The system is tested within a computer model of a water ballast tank, from a double sided cargo carrier, infused with corrosion areas. The initial results indicate that the UAV is able to successfully maneuver through tight entrance ways to explore and map the different compartments of a ballast tank while observing a higher percentage of corrosion than “vanilla” next best view.