Driverless Excavators: Inside Bedrock Robotics’ Autonomous Construction Sites
Bedrock Robotics has excavators doing productive work on live commercial job sites with nobody in the cab. I spent a day at the company’s test site outside Austin with the engineers who built it, a 30-year operator whose old seat is now empty, and the contractor paying for it. The safety case is the strongest part of the pitch, and the contractors using it talk about bidding more work rather than cutting crews.
There is a machine on nearly every commercial job site in America that takes about five years to learn properly. Five years before anyone hands it over and walks away. That number is the quiet reason construction’s labor conversation never resolves, because you cannot fix a five-year pipeline with a hiring push in March.
The demographics are not subtle either. NCCER projects around 41% of the US construction workforce will retire by 2031, and ABC puts the 2026 requirement at roughly 349,000 net new workers just to stay level, most of that covering people leaving rather than new demand. Set that against data centers, chip fabs and reshored manufacturing all breaking ground at once, and you land on the question every earthwork contractor is asking out loud: who runs the machines?
Bedrock Robotics has an answer that sounded like a pitch deck until this year. I wanted to see it rather than read about it, so I went to Proto Town, a sprawling site outside Austin built for the single purpose of teaching heavy machinery to run with no one inside.
Why Bedrock Started With The Hardest Machine On Site
Solve the excavator and the rest of the yard gets easier
Kevin Peterson, Bedrock’s co-founder and CTO, spent much of the last two decades on self-driving vehicles, most recently leading perception for Waymo’s trucking arm. He could have pointed the company almost anywhere, and he picked the machine nobody wants to start with. An excavator has six degrees of freedom plus a giant arm, Peterson calls it the quarterback of the job site, and it is brutally hard to learn, which is where that four-to-five-year figure comes from.
His logic is that the worst case forces you to build something scalable, so a dozer, loader or dump truck arrives later as a simpler version of a problem already beaten. The training approach will sound familiar to anyone following AI: large volumes of behavior data, then hard evaluation against it, much the way you would train a language model. Bedrock does not manufacture excavators either. It fits sensors and compute to machines contractors already own.
If we can do excavators, we can do any machine in construction.Kevin Peterson, co-founder and CTO, Bedrock Robotics
The Safety Test I Walked Into Myself
A 360 degree sensor suite against an operator’s 180 degree blind spot
Every machine carries cameras covering the full circle around it, plus lidar accurate to about a centimeter. It detects people, it detects vehicles, and works out the physics itself. Peterson’s team came from a world of toddlers stepping out from behind parked cars, so the target is measurably better than a human rather than level with one. I asked for a live, unrehearsed demonstration of the intrusion stop, and he agreed. Walking toward the cab of a 50-ton machine with nobody inside it, every instinct I have said turn around. Nobody touched a button. It saw us and it stopped.
That matters, because construction’s safety numbers have barely moved. BLS recorded 1,034 construction deaths in 2024, close to one in five of all US workplace fatalities, at a fatal injury rate of 9.2 per 100,000 full-time workers against an all-industry average of 3.3. Taking a person out of a cab that swings several tons through a blind arc removes a category of risk outright, and the contractor buying this told me the safety features alone would justify the spend.
The Operator Whose Seat It Used To Be
Thirty years in the cab, now setting up the computer that runs it
Steven Stapleton left the army in 1994, started as a laborer with a wet utility contractor out of Colorado, and worked up through operator to foreman across roughly 30 years around excavators. He now works at Bedrock. Asked about the workforce, his answer was blunt. It is bleak out there, few people are left who want to fill the seats, and he had wondered for years who would take their spots once his generation reached the rocking chair.
His reaction to watching a machine dig and load a truck with nobody inside was not what I expected from someone in his position. He said he felt relieved. Somebody still fuels it and checks the oils, so the physical presence has not gone. What shifted is where his day happens, setting up and monitoring an excavator in the field rather than sitting in a seat for ten hours.
It means more time with family, being able to enjoy other things and not have to worry about work all the time.Steven Stapleton, heavy equipment operator, Bedrock Robotics
The Customer Who Started Out Unimpressed
Champion Site Prep went from scoffing to buying inside a year
Trey Taparauskas is President and CEO of Champion Site Prep, a Central Texas earthwork contractor digging since 1985 and now running around 300 people. Told that some guys from Silicon Valley were coming to Texas to explain how they would change construction, he was not expecting much. He came away impressed. His team took longer, and what came back was some version of what is in it for us, and that is five to ten years out. He disagreed, told them Champion was going autonomous in 2026, and they came with him.
Champion ran supervised autonomy first, with a Bedrock employee in the cab, and more recently a machine worked one of its sites for several hours with nobody inside. Around 90% of the original skeptics have come around, by his estimate. On the question everyone wants answered he was direct: it does take a person out of that seat, but if that person is one of his best, Champion still wants them running a crew or managing a group of these machines. Without Bedrock, he told me, Champion would not be at 300 employees.
- Bid contingency. Earthwork carries one of the biggest variances on a project. Peterson’s example was 4 or 5 percent coming down to 2.
- Finance cost. A developer carrying a project at 8 to 9 percent interest can find that two months of schedule compression decides whether it works at all.
- Win rate. Bedrock’s partners report taking hard-bid projects by half a percent, which puts that contingency swing in perspective.
- Labor predictability. A labor cost that will not fluctuate, on a machine that cannot pick up an injury or take a day off.
What Comes After One Machine Digging
Teams of machines, orchestrated across a whole site
Ask Peterson where this goes in five years and the answer is fleets rather than units, with machines coordinating across a site instead of being driven one by one. Further out, Bedrock wants a foreman to explain verbally what needs to happen to a stockpile and have the machines sort out the rest. The partner list already spans the industry, from Champion at the subcontractor end, through Sundt and Zachry Construction on heavy civil and data center work, up to top 10 to 15 nationwide GCs.
So here is the practical version for anyone running dirt. The near-term question is what a driverless machine does to your contingency line, and whether your estimators have modeled it yet. Champion did that math early and won work it could not otherwise staff. Peterson argues that cheaper, more predictable earthwork makes more projects pencil out, which creates more work rather than less.
Autonomy, workforce, and the economics underneath both
The weekly briefing for the people making these calls, not the ones commenting on them.
Join 3000+ ReadersWhere Heavy Equipment Autonomy Actually Sits Today
Five stages, and what each one changes for a contractor
| Stage | Who is in the cab | What the contractor gets | Status in 2026 |
|---|---|---|---|
| Manual operation | A trained operator, 4 to 5 years to competence | Full flexibility, capacity capped by headcount | Still the bulk of the market |
| Machine control and GPS grade | Operator, assisted by the system | Better accuracy, less rework, same labor requirement | Widely adopted |
| Supervised autonomy | A Bedrock employee, watching and ready to take over | Proof the system holds up on real dirt, labor unchanged | Ran through 2025 and into 2026 |
| Driverless, operator out | Nobody | Hours that do not depend on staffing, predictable labor cost | Live commercial sites in Texas and Nevada |
| Coordinated fleets | Nobody, one supervisor across many machines | The whole earthwork phase replanned around machine availability | Bedrock’s stated next chapter |
Peterson’s definition on camera was specific: nobody in the cab, the machine completely self-sufficient, and nobody on the site watching it either. That last clause is what separates this from supervised autonomy, where a person sits in the seat while the hours accumulate. Bedrock describes it as the first fully autonomous excavator work completed on live commercial construction sites with an empty cab. (Source)
Real projects with owners, schedules and money attached, including a water treatment facility in Nevada and major earthworks in Texas. Taparauskas independently confirmed that a machine ran autonomously for several hours on a Champion Site Prep site with nobody inside. Proto Town in Lockhart, where the documentary was filmed, is Bedrock’s own development site rather than a customer project. (Source)
No. The Bedrock Operator is a retrofit that mounts to machines a contractor already owns, combining cameras, lidar, GPS and onboard compute, and it has been fitted to excavators from 20 to 80 tons across several major manufacturers. Because nothing is permanently modified, the machine can return to manual operation or be resold without a problem. That matters commercially, since it keeps the decision away from the capital committee. (Source)
The machine builds a live picture from a 360 degree camera and lidar suite accurate to roughly a centimeter, detects people and vehicles, and halts automatically when something gets too close. I tested that myself by walking toward a running machine, and it stopped without anyone intervening. For context on why this matters, construction recorded a fatal injury rate of 9.2 per 100,000 full-time workers in 2024 against an all-industry average of 3.3. (Source)
Both the operator and the contractor I spoke to described it as a change of role rather than a removal. Stapleton moved from sitting in a seat to setting up and monitoring machines in the field, and framed it around getting time back. Taparauskas was clear that a person comes out of that seat, and equally clear that his best people move into foreman, supervisor or fleet management positions. The backdrop is an industry needing roughly 349,000 net new workers in 2026, most of that to cover retirements. (Source)
Because it is the hardest machine on site, with six degrees of freedom and an arm that has to read ground conditions continuously. Peterson’s argument is that solving the worst case forces a scalable approach, and every machine tackled afterwards, from dozers and loaders to dump trucks, scrapers and compactors, benefits from what was learned first. Bedrock has said bulldozers and loaders are on the roadmap. (Source)
The company was founded in 2024 by Boris Sofman as CEO, Kevin Peterson as CTO, and Ajay Gummalla and Tom Eliaz as VPs of engineering. Three of the four came from Waymo, where Sofman led autonomous trucking and Peterson led perception for that division. Bedrock has raised $350 million to date, including a $270 million Series B in February 2026 that valued the company at $1.75 billion. (Source)
Related Articles
Bedrock Robotics Says Its Excavators Are Now Digging With Nobody in the Cab
41% of Your Workforce Retires by 2031 and Nobody’s Writing It Down
Construction Robotics in 2026: 4 Workflows Where Robots Are Actually Working
The Robot Report, Bedrock Robotics Announces Autonomous Excavation Milestone
PR Newswire, Bedrock Robotics Raises $270 Million in Series B Funding
Forbes, Waymo Vets Are Automating Construction Sites With Self-Driving Dirt Diggers
Inc, The AI Boom Has a Dirt Problem
BLS, Census of Fatal Occupational Injuries 2024
Westside Construction Group, ABC and AGC Workforce Data 2026
