Caterpillar Cat Command: the largest driverless truck fleet on earth, and it hauls rock

Every “first driverless” claim on this site — and we have reviewed fifty-five vehicles now — comes with an unspoken footnote: first on a public road. Drop that footnote and the league table inverts. The company with the most driverless trucks, the most driverless kilometres and the longest unbroken safety record is not Waymo, not Aurora, not Tesla. It is Caterpillar, and it has been quietly running trucks with nobody aboard since 2013.

On 14 September 2026, Caterpillar and the Virginia aggregates producer Luck Stone announced they were extending autonomous hauling to two more quarries, Boscobel and Bealeton, using Cat 775 trucks — the first time that model has been run autonomously anywhere. In the same release Caterpillar published the running total for its haulage autonomy: 13 billion tonnes moved and more than 455 million kilometres travelled, with zero reported injuries.

Those numbers are real, they come from Caterpillar's own newsroom rather than an analyst deck, and they are almost always cited badly. So let us do the comparison properly, because the honest version is more interesting than the lazy one.

The numbers, and what they are not

In 2022, when the fleet passed five billion tonnes, Caterpillar's group president for Resource Industries, Denise Johnson, said its trucks had “safely traveled nearly 200 million km … more than twice the experience in autonomous operations of any automobile manufacturer”. That was defensible at the time: Waymo did not announce its first million rider-only miles until February 2023.

Four years later the shape of the claim has changed. Caterpillar added roughly 255 million km in those four years. Waymo went from one million miles to two hundred million in three and a half. One curve is linear and industrial; the other is exponential and urban. Caterpillar is still in front on raw distance — and will be for a while — but it is no longer in a different league, and it is the only one of the two whose lead is shrinking.

That is before the part that matters. Caterpillar's 455 million kilometres were driven on private property: a bounded site, with no oncoming traffic, no pedestrians who have not been briefed, no cyclists, no children, no traffic lights, no unprotected left turn across a stream of strangers, and a map the operator is free to redraw at will. Every human being inside the geofence is an employee in high-visibility clothing who has been trained on what the trucks do. The trucks are centrally dispatched and their top speed, loaded, is 65.9 km/h.

Nobody at Caterpillar pretends otherwise. The company's own language is about removing operators “from hazardous or remote sites”, which is a safety argument about the place, not a claim to be a better driver than a person. It is worth holding that distinction, because the rest of the industry has spent a decade blurring it.

What Luck Stone actually bought

The interesting deployment is not the 370-tonne monsters in the Pilbara. It is the quarry.

In November 2024 Caterpillar and Luck Stone put an autonomous Cat 777 to work at the Bull Run plant in Chantilly, Virginia. Caterpillar called it two firsts at once: the first deployment of autonomous haulage in the aggregates industry, and the first autonomous truck in the 100-ton (90-tonne) class — a step down from the mining fleet, which spans the 139- to 370-tonne classes. Charlie Luck, the chairman and chief executive of Luck Companies, called it “a major accomplishment”; Johnson said the collaboration had “accelerated our ability to scale our proven mining solution for this sector”.

By July 2025 the site had hauled a million tons autonomously. Within the first year it passed two million, with no reported safety injuries and, Caterpillar says, productivity matching that of staffed machines “shortly after launch”. Eighteen months in, the figure was 3.5 million tons. In September 2026 Luck Stone signed up two more quarries and a smaller truck.

Read that sequence as a procurement decision rather than a technology story and it is the most straightforward business case for autonomy anywhere on this site. A quarry owner bought a proven system, on a machine it already runs, from a dealer it already uses — Carter Machinery, the local Cat dealer — to do a job it already understood, on land it owns. No regulator had to be persuaded. No statute had to be written. Nothing needed to be called a robotaxi.

The 775: a 64-tonne-class truck, and the smallest machine Caterpillar has yet put under Command for hauling — announced on 14 September 2026 for two more Luck Stone quarries in Virginia. The autonomy is not really on the truck anyway. MineStar assigns the loads, tracks the machines and holds the site map; the truck executes. The direction of travel is downwards in size and outwards from mining. Source: Caterpillar

How it works, and how little Caterpillar will say

Here is the oddest thing about the largest driverless fleet in the world: its maker publishes almost nothing about how it perceives.

Cat MineStar Command for hauling combines, in Caterpillar's words, “an advanced onboard perception system and proximity awareness” to detect obstacles, and will “automatically stop or slow the truck’s speed” when it finds one. It maintains following distances and speeds, assigns loads, tracks equipment, allows the site map to be redrawn on demand, and lets autonomous trucks “interact safely with staffed equipment and light vehicles”. That is the level of detail on offer. No sensor count. No range figures. No compute platform. No published safety framework of the kind Waymo puts out, and nothing like the parts list Xiaomi printed in its launch material for a car that does hands-on Level 2.

There is a benign reading and a less benign one. The benign: Caterpillar sells to mining engineers who evaluate on site trials and uptime, not to consumers who read spec sheets, and a mine's perception problem is genuinely narrower than a city's, so there is less to boast about. The less benign: an eleven-year safety record with no public incident taxonomy, no equivalent of an NHTSA recall filing and no independent conformity assessment of the kind Gatik commissioned from TÜV SÜD is a record that nobody outside Caterpillar and its customers can audit. “Zero reported injuries” is a strong claim. It is also self-reported, and the reporting regime on private industrial land is not the one that applies on a public road.

What Caterpillar does publish is the model list, and it reads like a fleet-management document rather than a product line: 789D, 793D, 793F, 797F, 794 AC, 798 AC, a 789D water truck, the 777, now the 775, with the 785 arriving “Command-ready” with part of the kit fitted at the factory. Retrofit kits are available for machines already in service. Autonomy here is an option box, not a model.

The direction of travel is towards the quarry, not away from it

The reason this review sits next to our long-haul coverage rather than in a corner marked “industrial” is that the public-road companies keep walking towards Caterpillar's ground, not away from it.

Kodiak is the clearest case. It spent years on interstate freight and then went driverless for real in the Permian Basin, hauling sand for Atlas Energy on private lease roads — a bounded site, a repeated route, an industrial customer who controls both ends. Driverless first where the road is owned; public highways after. Gatik reached the same conclusion from the other direction, building a business out of fixed middle-mile routes for customers who own the distribution centre and the store. Einride put a cabless vehicle inside industrial sites and a human in a remote seat.

Set against those, Aurora's open-highway programme — roughly 440,000 driverless miles by June 2026, ten Sun Belt routes, a target of 200 driverless trucks by the end of this year and thirty thousand by 2030 — is the ambitious version, and it is the one still furthest from the economics. The pattern across all of them is the same: driverless arrives first where somebody owns the road. Caterpillar got there in 2013 because mining already owned it.

Which raises the question the quarry answers. Bull Run is not a mine. It is a commercial aggregates site twenty-five miles from Washington DC, with contractors, deliveries and haulage trucks coming and going. It is a step from the most controlled environment in heavy industry towards something a little more like the world. The 775 expansion says Luck Stone found that step worth taking twice more.

The Nvidia footnote that says the most

On 7 January 2026 Caterpillar and Nvidia announced an expanded collaboration: Jetson Thor, Riva speech models, Omniverse digital twins of factories, and “AI-driven autonomy capabilities” for construction and mining equipment. Joe Creed, Caterpillar's chief executive, framed it as leading “with advanced technology in our machines”; Jensen Huang said the two were partnering “from autonomous construction fleets to the AI data centers”.

The headline deliverable, debuted at CES, is a voice-activated in-cab AI assistant giving operators coaching and safety alerts. Command for hauling is not mentioned in the announcement at all.

Sit with that for a moment. Jetson Thor is the same silicon family as the 700-TOPS Thor chip that Xiaomi fits to every YU7 in China to run a car that is legally hands-on. At Caterpillar, that generation of compute is going into the cab of a machine that still has a human in it — while the trucks with nobody aboard carry a perception stack the company will not describe, and evidently do not need the upgrade. The most autonomous vehicles in commercial service are running on the least fashionable hardware, and it is working.

The comparison that is fair

Autonomy is usually scored as a single ladder, and this site publishes a scale that does exactly that. Caterpillar breaks it, because it scores maximum on the thing the ladder measures — no human in or supervising the vehicle, in commercial service, for years — while operating in conditions the ladder assumes away.

The right way to hold both facts at once is this. What Caterpillar demonstrates is not that driving is solved. It is that the hard part of autonomy was never the vehicle. Give a competent engineering organisation a bounded site, a known map, a trained population and a single owner of the road, and driverless operation has been a solved commercial problem for over a decade. Everything the public-road industry is still struggling with — the pedestrian who steps out, the closed lane nobody signposted, the liability question, the city that has not consented — is a property of the road, not of the truck.

That is worth saying plainly at a moment when the fastest, simplest-looking road on the network has just turned out to be the one that most embarrassed the most advanced robotaxi fleet in the world.