Towards robotic exploration without external infrastructure in underground mining environments: a case study from the PERSEPHONE project

The PERSEPHONE Project supports the EU’s strategy to access deeper, previously abandoned or otherwise challenging underground mineral deposits in a more sustainable, safe and digitalised manner. In this context, the field deployment reported here at the Koutzi Mine (Evia, Greece) in September 2025 represents one of the demonstration missions of PERSEPHONE, during which a robotic platform performed mapping, relocalisation and multispectral mineral imaging without reliance on external infrastructure.

Robotic exploration of underground environments can serve not only as a means of new discovery, but also as a valuable tool for the remapping of historic galleries and more broadly for subterranean exploration (including caves and other naturally occurring voids). For instance, the UNEXMIN/UNEXUP projects have employed robotic systems to re-survey Europe’s abandoned flooded mines, as well natural flooded cavities like  the Molnár János cave (Hungary).

The geological setting of the Koutzi Mine is characterised by a narrow-vein magnesite deposit hosted in ophiolitic ultramafic lithologies on the island of Evia. This historic mine was reopened in 2021 and employs sub-level stoping with battery-operated excavators, reflecting a precision extraction philosophy designed to minimise environmental footprint. However, some of the older, smaller galleries remain unsafe for human exploration. The occurrence of magnesite (MgCO₃), frequently resulting from carbonation of ultramafic rocks, together with accessory white minerals such as sepiolite or opal in fault or alteration zones, provides a good target for multispectral imaging: determining vein type, thickness and mineral differentiation in this environment improves both exploration efficiency and robotics mission planning.

The exploration campaign comprised two phases. In the first phase, a agile mobile robot equipped with LiDAR and IMU sensors operated autonomously within the gallery, constructing a detailed volumetric map of several sections of the mine without use of GPS or pre-deployed reference beacons. Zones of interest were identified using the onboard visible-light camera to locate white-mineral zones. In the second phase, a second robot was introduced, successfully relocalized itself within the map created by the first robot and deployed to capture high-quality multispectral imaging of the identified white-mineral vein zones. The multispectral imaging subsystem comprised a near-infrared (NIR) camera and a UV-fluorescence camera mounted on the robot’s sensor suite. The objective was to acquire precise spectral–spatial data on vein geometries and white-mineral occurrences (distinguishing magnesite, sepiolite and opal) and to characterize thickness and orientation of the mineralized zones. By planning reference viewpoints with high overlap (80 %), the system links multispectral data with the 3D map context and supports subsequent data-driven analytics. Together with autonomous mapping and relocalization in absence of external infrastructure, this experiment provides a proof-of-concept of integrated robotic exploration, targeted mineral sensing and operational autonomy in an underground mining environment.