Cave Exploration Robot: Survey Tapes, SubT and ABB

Caves are dark, wet and full of tight squeezes, and for centuries every map of one was drawn by people crawling through it with a tape and a compass. A cave exploration robot does the same job with cameras, depth sensors and software that works out where the robot is as it goes. This post follows the work from hand surveys and the first digital instruments to the mine-mapping robots of the 2000s and DARPA's Subterranean Challenge, then looks at this autumn's news from a Spanish cave, US mines and the Moon, and at where ABB robots could fit. Rubble crawlers for disasters are a separate topic, covered in our post on search and rescue robots.

Light falls through the ceiling of a wide lava tube passage at Craters of the Moon, 2022. Photo: NPS via Wikimedia Commons (public domain)

How cavers mapped caves with tape, compass and pencil

The first known cave plan dates from 1546 and shows the man-made Stufe di Nerone near Naples. The Baumannshöhle in Germany became the first natural cave to be mapped, in a sketch from 1656, and the Grotte de Miremont in France, surveyed by a civil engineer in 1765, is probably the first cave measured accurately with instruments. Édouard-Alfred Martel was the first person to describe surveying techniques. The basic method is still in use. Starting from a fixed point such as the entrance, cavers measure each leg between stations with a compass for direction, a clinometer for slope and a tape for distance, and sketch the passage as they go.

A caver with a survey booklet strapped to her helmet crawls through a tight passage while surveying at Wind Cave, 2010. Photo: NPS via Wikimedia Commons (public domain)

Computers joined in 1964, when cavers plotted Ireland's Fergus River Cave, the first cave whose centreline was calculated by a computer. Digital instruments began changing surveys in the late 1990s. In the 2000s a Swiss caver added inclination and direction sensors to a commercial laser rangefinder, which produced the DistoX2, a device that records a whole survey leg in one laser shot. Mapping robots grew out of mining. In July 2002 nine miners at Quecreek, Pennsylvania, were trapped for over 77 hours after breaking into a flooded old mine that an undated, uncertified map had not shown properly, and the accident led roboticist William L. Whittaker to develop robots that could map collapsed or abandoned mines.

Cave surveying instruments, from top: a DistoX2 and SAP2, SAP5, SAP6 and BRIC5 devices, 2024. Photo: Cloud200 / Wikimedia Commons (CC BY-SA 4.0)

DARPA's Subterranean Challenge, announced in early 2018, sent robot teams into human-made tunnels, the urban underground and natural caves. At the Tunnel Circuit in August 2019, held in NIOSH's experimental coal mine near Pittsburgh, many teams lost contact with their robots after the first bend, and the Cave Circuit planned for autumn 2020 was cancelled because of COVID-19. The final, in September 2021, ran through a kilometre-long course in the Louisville Mega Cavern with 40 artifacts to find in an hour. CERBERUS, whose robots included ANYbotics' ANYmal walker, and CSIRO Data61 both found 23. CERBERUS reached its last find 46 seconds sooner and took the $2 million prize.

A Spot quadruped marked CSIRO and fitted with a sensor payload, with team members underground during DARPA's Subterranean Challenge. Photo: DARPA

Spanish caves, US mines and drones in the dark

CSIRO Data61's robots built a map of that final course that differed from DARPA's ground truth by less than 1 percent, close to what a team of expert humans needed many days to produce. Wheeled robots gave the most reliable mobility, quadrupeds proved surprisingly capable on rough ground, and drones explored some of the larger caverns. Natural caves are still difficult, with total darkness, complex shapes, narrow passages and damp, reflective walls. In October 2026 researchers from the University of Malaga and the Polytechnic University of Turin released Cavers, a 335 GB dataset recorded in the Cueva de la Victoria with a rover carrying thermal cameras, depth cameras and inertial sensors, checked against a motion capture system installed in the cave.

An ANYmal legged robot from ANYbotics, built for industrial inspection, 2022. Photo: ANYbotics / Wikimedia Commons (CC BY-SA 4.0)

In mining, these robots already work commercially. Exyn Technologies, whose ExynAI software lets aerial and ground robots navigate and map places with no GPS and no prior map, said on 19 October 2026 that it had joined MAP-POD, an Oak Ridge led project chosen under the US Department of Energy's $29.5 million Mine of the Future initiative. At Eldorado Gold's Tüprag mine in Turkey, Exyn says its Nexys system cut survey capture from about 40 minutes to about 5 minutes per scan. Drones help above ground as well. On Australia's Nullarbor Plain, a team led by Curtin University's Matej Lipar used satellite radar and drones to reach cave openings in cliffs above the Great Australian Bight and found the cave systems much larger than thought.

A Flyability Elios 2 caged inspection drone on display at Embry-Riddle Aeronautical University, 2025. Photo: ZLEA / Wikimedia Commons (CC BY-SA 4.0)

ABB does not make cave exploration robots, so any role for it there is a possibility. It does have a robot working at the rock face. The ABB Robot Charger, tested with Boliden and LKAB at Boliden's Garpenberg zinc mine in Sweden, uses a vision system to find drilled holes and fill them with explosive charges, so people no longer have to stand near the unsupported rock while it works. We covered it in our post on robotic blast hole charging. In factories, ABB's Flexley Mover P604 handles the mapping side: its 3D Visual SLAM lets each robot build its own map of its workspace, accurate to within 10 mm, and share it with the rest of the fleet.

The ABB Robot Charger finds drilled holes in a rock face and installs explosive charges. Photo: ABB

Moon pits, mineral hunts and what still stops cave robots

Mapping robots are now expected to find things as well as draw maps. The DOE initiative funds 17 national laboratory projects, and MAP-POD aims to combine autonomous robots, sensors and AI to map underground workings and identify minerals in real time. DARPA's program manager Tim Chung described the scale of the job to IEEE Spectrum: "if you look at the amount of infrastructure that we've built underground, it's just massive." Whether vision and SLAM tools proven in factories, such as the Flexley Mover's, could one day guide ABB machines through mines is an open question, and ABB has announced no such product.

An ABB Flexley Mover P604 among wooden pallets; it finds its way with 3D Visual SLAM. Photo: ABB

NASA is now looking at caves on the Moon. On 30 September 2026 it selected GIMLI, a payload that will land near the Marius Hills Pit, a hole about 90 m across and 50 m deep that may open into a lava tube; it is one of about 300 pits known on the Moon. GIMLI will not go in. It will probe the ground with radar, a gravimeter and seismic sensors. Closer to home, cave maps matter to ordinary travellers. Drivers crossing the Nullarbor pass over collapse trenches without noticing, and Lipar warns that "something that hasn't collapsed yet is more dangerous."

The Marius Hills Pit on the Moon, a possible lava tube skylight, seen by NASA's Lunar Reconnaissance Orbiter, with a close view on the right. Photo: NASA/LROC/Arizona State University (public domain)

Communication is the first open problem. Radio does not travel far underground, so SubT teams ran tethers or had robots drop mesh network nodes like breadcrumbs, and once contact is lost the robots must decide for themselves where to go. Darkness, dust, damp reflective rock and rockfalls confuse sensors, and most of the research robots at SubT still needed careful attention from their human teammates, even though only one person per team was allowed to interact with them during a run. Caves also need protecting. Lipar calls them natural archives, and a robot sent to map one has to leave it as it found it.

The ABB Robot Charger arm close to the rock face, where drilled holes are marked in green. Photo: ABB

Innovation

AI solutions for effortless ABB robot control.

Automation

Robotics

ceojohntran@chatwithrobot.net

+84905311611

© 2025. All rights reserved.

qtran1215@gmail.com