Towards automatic recognition of mining targets using an autonomous robot J. Quintana 1 , R.Garcia 1 , L. Neumann 2 , R. Campos 1 , T. Weiss 3 , K. Köser 3 , J. Mohrmann 3 , J. Greinert 3,4 1 Coronis Computing SL, 2 ICREA, 3 GEOMAR, 4 University of Kiel josepqp@gmail.com, rafael.garcia100@gmail.com, lneumann@eia.udg.edu, ricardcd@gmail.com, {tweiss, kkoeser, jmohrmann, jgreinert}@geomar.de Abstract—Modern technology like cell phones, wind power plants or electric cars require resources such as certain metals or rare earth elements with limited deposits on land, or expensive or difficult to obtain. Consequently, resources in the oceans like polymetallic nodules, massive sulfides and cobalt crusts are becoming more and more interesting for mining companies. Since mining in the deep sea needs careful consideration and mapping of the concerned locations, might require ecological compensation areas, and is a huge endeavor with enormous costs, logistics and machinery, detailed exploration and spatial planning, resource quantification and environment mapping are inevitable steps early in the process. While traditionally, experts performed several manual steps of map creation, interpretation, target localization, sampling and resource estimation, this paper describes a new pipeline for manganese nodule detection combining acoustic and visual information, that is ultimately intended to run automatically on an Autonomous Underwater Vehicle (AUV) without any user interaction. Keywords—Automatic recognition, mining targets, AUV I. INTRODUCTION Large amounts of Manganese (Mn) nodules have been found at several places in water depths of 4000m to 6000m, e.g. in the Clarion-Clipperton zone, the Peru Basin, the Penrhyn Basin and the Indian Ocean, but the significant water depth makes their detection, analysis and quantification a difficult task [1]. Nodule size typically varies between 3 and 10 cm. They contain mostly Mn, but also iron, nickel, copper, titanium, cobalt or other materials [2]. Among the different regions, big differences exist in the nodule coverage area (from 3000 km 2 to 9 million km 2 ), amount of nodules per area (5 kg/m 2 of Mn-nodules to 75 kg/m 2 ) and nodule composition (i.e. contained elements and percentage) [3]. Consequently, for assessing the resources available in a particular region, nodule fields have to be found and their parameters such as nodule composition and their distribution on the seafloor are required. To allow for sustainable deep-sea mining once a promising location has been identified, a good understanding of the eco-system is mandatory, requiring further environmental baseline studies as well as ecological impact assessments [4]. Figure 1: Manganese nodule field (photo: ROV-Team GEOMAR) The remainder of this paper will focus on the first part, i.e. finding promising regions that can be characterized. Towards this end, the H2020 EU project ROBUST aims at a new automated methodology for finding Mn-nodules using a novel AUV that performs these steps automatically:(1) acoustic mapping to localize potential Mn areas, (2) exploration and visual identification of the Mn-nodules and (3) automated sampling to retrieve nodule composition. All of these steps should ideally be carried out autonomously by a robot, with no user interaction required. Towards this goal, this paper presents processing for steps (1)+(2) that are demonstrated offline on real-world data. The acoustic mapping is carried out using an AUV following a predefined lawn mower trajectory to obtain a first map of a large area covering square kilometers. Using this map and extracting relevant terrain and backscatter features according to the models learned in [5], the system identifies zones with high probability of containing Mn-nodules. After having identified a working area with high Mn- nodule probability within the acoustic maps, the AUV will navigate closer to the seafloor in order to optically detect the potential nodules using a downward-looking camera. During this step, the robot would detect and identify any individual object on the sediment that resembles a manganese nodule. Once a candidate nodule is detected, the system estimates its 3D position and size. The AUV will then perform an operational inspection on the nodule, i.e. in-situ analysis using laser-induced breakdown spectroscopy (LIBS) [6]. The LIBS reading