In the western Australian region of the Pilbara, an area of red dirt roughly the size of California, three companies — Rio Tinto, BHP, and Fortescue — operate the most heavily automated heavy-industrial complex on the surface of the Earth. Rio Tinto alone runs an autonomous haul truck fleet across five of its 18 Pilbara iron ore mines, with roughly a quarter of the company’s 400-truck fleet operating without drivers in 2025 and a retrofit program adding 48 more Komatsu and Caterpillar trucks to autonomous operations. The Cat 793F and 797F haul trucks involved are 380-ton machines whose tires are 13 feet tall and whose cabs sit 24 feet above the ground; the trucks drive themselves up and down haul roads using GPS, lidar, and a centralized fleet-management system in a control room in Perth, 1,500 kilometers away. Twenty-five percent of all material moved by Rio Tinto across the Pilbara in any given year is moved by a robot.
The autonomous freight railway that ships the resulting iron ore from those mines to the export ports of Dampier and Cape Lambert — Rio Tinto’s AutoHaul system, fully driverless since 2019 — is, by a substantial margin, the largest autonomous robot on Earth. Each AutoHaul train is up to 2.4 kilometers long, weighs roughly 38,000 tonnes loaded, consists of 240 locomotives and 16,500 ore cars across the fleet, and operates with no human onboard the train itself. The trains move iron ore over 1,700 kilometers of track at speeds up to 80 kilometers per hour. They are monitored from the Perth Operations Centre. Their reliability is higher than the human-operated trains they replaced. None of this gets the coverage that a humanoid robot doing a backflip gets. All of it has been operating commercially since the year before the first Boston Dynamics Spot shipped to its first paying customer.
This is the part of the robotics industry that the consumer press doesn’t cover, that the venture capital community doesn’t fund, and that the companies generating the humanoid robot headlines are not, with rare exception, the same companies producing. Mining automation is the success story the robotics industry has, almost without exception, refused to tell about itself.
Surface mining and the autonomous haul truck
The autonomous haul truck industry is dominated by two manufacturers — Caterpillar and Komatsu — and almost entirely by two customers: Rio Tinto and BHP, with Fortescue Metals Group as a fast-growing third. Caterpillar’s autonomous fleet — operating under the Command for Hauling system — has moved more than 6.6 billion tonnes of material since the system was first commercialized in 1991. The Komatsu FrontRunner Autonomous Haulage System (AHS) has been operating at Rio Tinto’s West Angelas mine in the Pilbara since 2008, making the iron ore industry the longest continuously operating autonomous heavy-vehicle deployment in any industry, anywhere. By comparison, Waymo’s first commercial robotaxi service in Phoenix did not launch until 2018.
The economic argument for mining automation is brutal in its simplicity. An autonomous haul truck runs roughly 700 more hours per year than a human-operated equivalent because it does not require shift changes, lunch breaks, or rotation between drivers. The unit cost of moving a tonne of iron ore drops by roughly 15 percent. The accident rate drops by more, in an industry where the historical fatality rate is well above the average for industrial work and where the hyper-specialized labor force lives in fly-in-fly-out worker camps with serious mental health and retention problems. Rio Tinto, BHP, and Fortescue did not build these autonomous fleets because robotics is fashionable. They built them because the alternative — manual operations across a multi-billion-tonne-per-year industrial process — is more expensive, more dangerous, and more difficult to staff. The same operational logic that made drone delivery economically rational for medical supplies in rural Rwanda made autonomous haul trucks economically rational for iron ore in the Pilbara. The difference is that the iron ore industry has been deploying the technology for 17 years.
The decarbonization wave nobody saw coming
The 2025-2026 inflection in mining automation is that the same autonomous fleets are now electrifying. On December 5, 2025, BHP and Rio Tinto jointly welcomed the first Cat 793 XE Early Learner battery-electric haul trucks to BHP’s Jimblebar iron ore mine in the Pilbara. The 793 XE is a 290-tonne payload battery-electric haul truck — the largest battery-electric vehicle ever commercially deployed in any industry. Two units arrived at Jimblebar for joint on-site testing between BHP, Rio Tinto, and Caterpillar, with operations expected to ramp to a scaled trial across multiple Pilbara mines through 2026. Six weeks earlier, on October 27, 2025, Rio Tinto launched a separate battery-electric trial at its Oyu Tolgoi copper mine in Mongolia — eight 91-tonne Tonly trucks built by China’s State Power Investment Corporation Qiyuan, paired with 13 800-kWh batteries that can be swapped in less than seven minutes at a dedicated swap station. The Oyu Tolgoi fleet is Rio Tinto’s first commercial battery-electric mining deployment, and it is built on Chinese battery-swap technology rather than American or European designs.
Mining haulage accounts for roughly 30 to 50 percent of the diesel consumption at a major iron ore or copper operation, and is the largest single source of Scope 1 and Scope 2 emissions at the average mine. The electrification of haul trucks is therefore both the largest decarbonization lever available to the mining industry and the most operationally consequential — replacing a fleet that runs 24 hours per day, 365 days per year, in some of the most remote operating environments on Earth. The fact that the world’s three largest iron ore producers and the largest copper producer are simultaneously deploying battery-electric haul trucks in 2026, on two continents, with vehicle platforms supplied by both American and Chinese manufacturers, is the kind of structural industry shift that mining trade publications cover and that the general business press largely ignores. The trucks themselves are essentially the same battery-electric heavy-duty platform that the freight industry has been promising for a decade — except that the mining industry has actually deployed them, at commercial scale, under operating conditions that would destroy a standard highway truck.
Underground mining and the operator in the surface office
Underground mining is where the case for robotics is most acute. The accident rate in deep underground mining — copper, gold, nickel, uranium — is higher than in surface operations by every measurable category. Heat, dust, rock fall, ventilation failures, and methane buildup combine to make the underground environment one of the worst occupational settings in any industry. Removing humans from that environment is the single largest safety improvement available to the mining sector — and the operational obstacle is not whether the technology exists but whether the existing workforce can be persuaded to accept it.
Sandvik and Epiroc are the two manufacturers that dominate the underground autonomous equipment market. Sandvik’s AutoMine system has been operating since 2004 and currently runs autonomous load-haul-dump (LHD) machines, drill rigs, and truck fleets across more than 70 underground mines worldwide. Epiroc’s AutoNav system performs the equivalent function on its own LHDs and drill rigs. At Westgold Resources’ Big Bell mine in Western Australia, Epiroc AutoNav LHDs are being managed by operators sitting in an automation center on the surface of the mine, with Multiple Machine Control allowing a single operator to supervise multiple loaders simultaneously — moving roughly 30 additional buckets of material per 24-hour shift compared to manual operation, because the autonomous machines continue working during the cross-shift change and re-entry times when humans are required to evacuate. The mine doesn’t need to stop for shift changes. The robots don’t go home.
The supervisory model in underground mining — one operator, multiple autonomous machines, surface-based control room — is structurally identical to the supervisory model that healthcare robots have begun enabling in American hospitals, to the Norwegian aquaculture model where two technicians in Trondheim oversee 17 sea-cage installations, and to the autonomous haulage operations centers in Perth that monitor hundreds of Pilbara haul trucks across multiple mine sites. The work is no longer happening at the location of the work. The work is happening in a control room, and the location of the work is staffed by machines.
Robot dogs on the offshore rig
The oil and gas industry has, since roughly 2020, become the largest non-military commercial customer for quadruped robots. BP’s Mad Dog platform in the deepwater Gulf of Mexico has been operating Boston Dynamics’ Spot since 2020 — reading gauges, identifying corrosion, scanning for thermal anomalies, and carrying methane-detection payloads on autonomous patrol rounds that previously required a human technician to walk the same route in full PPE. Shell’s Energy and Chemicals Park Pernis in Rotterdam — the largest oil refinery in the European Union — operates a mixed fleet of Spot, ANYbotics ANYmal X, tracked inspection robots, and aerial drones that conduct continuous autonomous inspections across the entire facility, with the data feeding into Shell’s enterprise asset-management software and the fleet supervised by technicians who can remotely control any single robot from a gamepad. Petrobras has deployed ANYmal robots at its onshore refineries and on its FPSO production vessels off the Brazilian coast. Petronas — Malaysia’s state-owned oil company — has run ANYmal trials at both onshore and offshore facilities since 2022, validating the platform’s performance under saltwater corrosion, tropical storms, and slippery offshore deck conditions.
The Swiss-based ANYbotics, spun out of ETH Zürich in 2016, has built its commercial business around oil and gas inspection in a way Boston Dynamics has not. The company’s ANYmal X is, as of 2025, the only quadruped robot certified for Zone 1 hazardous areas — environments where explosive gas mixtures are present continuously enough to require equipment certification under the ATEX and IECEx standards that govern offshore oil platforms. The 2026 release of the ANYmal XD — a larger, more rugged successor — is being timed to coincide with the renewable-energy industry’s push into offshore floating wind, where the same kind of platform inspection will be required at scale. Equinor has trialled ANYmal X at its Kårstø gas processing facility in Norway. Aker BP, Cognite, and ANYbotics have partnered on the Valhall platform in the North Sea — the world’s first attempt at fully remote inspection of an offshore production platform using autonomous quadrupeds. The structural argument for offshore robotic inspection is identical to the argument for autonomous haul trucks: the work is dangerous, the labor is expensive, the platforms operate 24/7, and the alternative is a human in a survival suit walking across a wet steel deck in 40-knot winds.
The methane detection drone and the regulatory inflection
In October 2025, the U.S. Environmental Protection Agency formally approved a category of autonomous methane-detection drones for OOOOa and OOOOb compliance — the EPA regulations that require oil and gas operators to detect and repair methane leaks across their production, gathering, and storage operations. The October 29, 2025 decision was the first time the agency authorized drone-based remote inspections as a substitute for manual leak detection and repair (LDAR) walking surveys. The approval shifts the economics of methane regulation: an autonomous drone equipped with a TDLAS (tunable diode laser absorption spectroscopy) sensor like the BLV Tech BL-CH4 can survey a pipeline corridor or compressor station at a small fraction of the cost of a human technician with a handheld sensor, and can do it weekly rather than annually.
The midstream pipeline industry — the long-distance natural gas and oil transportation network that runs across the rural United States — is the next frontier. The economics of drone-based pipeline inspection only work if a single operator can fly a drone hundreds of miles beyond visual line of sight (BVLOS) without continuously moving, which requires the FAA Part 108 BVLOS rulemaking that has been promised for the drone delivery industry since 2023. The Federal Aviation Administration’s BVLOS regulatory framework — published in proposed form in 2024 and expected to be finalized in 2026 — will simultaneously open commercial drone delivery, agricultural drone swarms, and oil and gas pipeline inspection to the kind of long-range autonomous flight that is currently allowed only under restricted experimental waivers. The same rulemaking that enables Zipline to drop a package at a Walmart cul-de-sac is the rulemaking that allows an oil and gas operator to fly a methane drone 200 miles along a buried pipeline without launching a chase vehicle. The economic logic is identical across industries. The regulatory bottleneck is identical. The technology is identical. The application labels are different.
Tailings dam monitoring and the Brumadinho effect
On January 25, 2019, a tailings storage dam at Vale’s Córrego do Feijão iron ore mine in Brumadinho, Brazil — a 720-meter-long, 86-meter-tall structure storing 12.37 million cubic meters of mining waste — collapsed without warning. The released slurry killed 272 people, including most of Vale’s on-site administrative workforce who were in the mine’s cafeteria at the time. The collapse remains the worst industrial accident in Brazilian history and the worst tailings dam failure on a measured-deaths basis since the Romans started building dams.
The Brumadinho disaster — combined with the 2015 Samarco Fundão failure that killed 19, the 2014 Mount Polley failure in Canada, and the 2022 Jagersfontein collapse in South Africa — restructured the global mining industry’s approach to tailings storage facility monitoring. The technology that did the restructuring was, in operational terms, drone-based ground-penetrating radar. Chilean mining companies now fly DJI M600 Pro platforms equipped with RadarTeam SE70 GPR sensors over their tailings dams on a monthly basis, generating high-resolution subsurface images that can detect humidity buildup inside the dam wall before it becomes structural liquefaction — the failure mode that destroyed the Brumadinho dam. Brazilian operators run continuous drone-based monitoring on every active tailings facility. Australian and Canadian operators have integrated tailings dam monitoring into the same fleet-management systems that operate the autonomous haul trucks. The technology is functionally similar to the variable-rate spraying drones now mapping every commercial soybean field in Brazil, and to the civil engineering monitoring drones covered in the Pipe Dreams cluster, and on every dam covered by the U.S. Army Corps of Engineers — but it took 272 deaths to make the case at scale.
The deep drilling and the resource frontier
Mining and oil and gas exploration are, structurally, the same engineering problem — get an industrial process into the ground, extract a valuable commodity, and bring it to the surface — separated by the temperature, depth, and chemistry of the target. The deepest current oil wells extend to roughly 12,289 meters of measured length, set by the Al Shaheen Oil Field’s BD-04A well in Qatar in May 2008. The deepest current scientific borehole is the Kola Superdeep at 12,262 meters of vertical depth, set in 1990 and unmatched since. The deepest current mining operation is the Mponeng gold mine in South Africa at approximately 4 kilometers below the surface, which is roughly a third of the Kola depth, and where temperatures at the working face reach 60 degrees Celsius and rock pressure measures in the hundreds of megapascals. Every deeper extraction operation — and the global mining industry has been pushing deeper as surface deposits deplete — requires the same family of autonomy, sensor, and remote-control technology that the petroleum industry has been developing for decades.
The 2026 inflection on the resource side is that the critical-minerals supply chain — copper, lithium, nickel, cobalt, the rare-earth metals, gallium, germanium, the uranium feedstock for the AI-data-center nuclear renaissance — is suddenly economically interesting to the same hyperscalers, sovereign-wealth funds, and federal industrial-policy programs that ignored mining for the last 30 years. The ethical questions around cobalt and the Congolese supply chain, around lithium and Argentine indigenous communities, around Chinese refining dominance in gallium and germanium — none of these get easier when the mining industry electrifies and automates. They get more economically consequential, because the volumes required to support a chip-driven AI economy and a fully electrified industrial base are larger than the volumes the mining industry has historically produced. The robotics is the means by which mining will respond to the volume demand. It is not the means by which mining will get less politically contested.
The Quaise option, and the bet that drilling cost can collapse
One last piece. Quaise Energy — the Houston-based MIT spin-out that has been developing millimeter-wave drilling technology that ablates rock using a gyrotron rather than a conventional drill bit — drilled 100 meters of Texas granite in a July 2025 field test, a record for the technology. Quaise’s bet is that the same gyrotron-based system that could potentially make deep geothermal drilling economically viable at depths of 20 kilometers will, by extension, make deep mining and deep oil exploration economically viable at depths and temperatures that conventional drilling cannot reach. If the technology works at commercial scale — and the engineering risk on that “if” is enormous — the global resource frontier will move from the depths the existing drilling industry can reach to the depths the next-generation drilling industry can reach, which is roughly twice as deep at twice the temperature. That is the same family of bet the autonomous-haulage industry made in 1991, and that the early offshore-platform-inspection robotics industry made in 2018. The technology took a decade to scale, but the case was built on the same logic: dangerous environment, expensive labor, continuous operation, and the alternative was getting worse every year.
What 2026 actually looks like across the mining and oil patch
Twenty-five percent of all iron ore moved across Rio Tinto’s Pilbara operations is being moved by an autonomous haul truck in 2026. The trucks are watched by a control room in Perth. The first 290-tonne battery-electric haul trucks have arrived at BHP’s Jimblebar mine. Eight Chinese battery-swap electric trucks are running at Rio Tinto’s Mongolian copper mine on 800-kilowatt-hour battery packs that swap in seven minutes. Underground autonomous LHDs at Westgold Resources’ Big Bell mine are being supervised from a surface office by a single operator managing multiple machines. BP’s Spot platforms are walking the deck of an offshore rig in the Gulf of Mexico, ANYbotics ANYmal X is the only quadruped certified for Zone 1 hazardous areas at Equinor and Aker BP’s North Sea facilities, and the ANYmal XD is set to ship in 2026 to expand the installed base of industrial quadrupeds beyond the few hundred currently in commercial service. The EPA has approved autonomous methane-detection drones for OOOOa and OOOOb compliance. Tailings dams across Brazil, Chile, Australia, and Canada are being monitored by drone-mounted ground-penetrating radar systems that did not exist before Brumadinho killed 272 people in January 2019. And the Pentagon, the AI hyperscalers, and the European Union’s industrial-policy apparatus are simultaneously realizing that the critical minerals required to power any of this — the lithium, the copper, the rare earths, the cobalt, the gallium, the germanium, the uranium — require an additional decade of investment in extraction infrastructure that has barely been started.
The autonomous mining truck is not a humanoid robot. It does not have a face. It does not pass the uncanny valley test because nobody designed it to. The autonomous ROV inspecting a subsea pipeline is not a humanoid robot. It does not interact with humans because there are no humans within 4,000 meters of its operating depth. The autonomous Spot patrol on the BP Mad Dog platform is, technically, a quadruped, and it is doing the work that the civilian humanoid manufacturers have been promising will be the killer application of their product for the last decade, except that Spot was already doing it in 2020. The work of mining, drilling, hauling, inspecting, and moving roughly 90 billion tonnes of material per year across the global resource economy is being done — quietly, in volume, in operating environments that no consumer will ever see — by a robot population that nobody in the consumer technology press covers, that the defense robotics community treats as adjacent technology rather than the main event, and that is, by every measurable metric, the most operationally mature deployment of industrial robotics on the planet. The Pilbara haul trucks moved more material in 2025 than the entire combined output of every humanoid robot factory on Earth, and they did it on hardware platforms that have been in continuous operation for longer than most of the consumer robotics companies have existed. The robots that matter most are, once again, the ones that do not look like robots — and the industry that built them was, once again, doing the work while the press was watching somebody else’s demo.