Most construction-robotics founders come from one of two places: software engineers who learned about dirt, or equipment people who learned to code. Ryan Luke Johns came from neither. He came from architecture — and that origin story turns out to explain a great deal about the company he now runs.
Johns is the co-founder and chief executive of Gravis Robotics, the Swiss firm turning conventional excavators and loaders into machines that can operate themselves. At CONEXPO-CON/AGG 2026 in Las Vegas, Gravis made its full commercial entry into the United States and walked away with the show’s Contractors’ Choice award for technology — a notable verdict, because that prize is voted on by the contractors themselves, the people whose skepticism is hardest to win. For Johns, the moment carried a thesis: “ConExpo 2026 marks the moment where Physical AI moves from the R&D lab to the American jobsite.”
To understand why he can say that with conviction, it helps to follow the unusual road that brought him there.
From Drawing Boards to Robot Arms
Johns studied architecture with a concentration in mathematics at Columbia, then took a Master of Architecture at Princeton. But his interest was never confined to designing buildings — it was in the act of making, and specifically in how machines and materials could be brought into a more intelligent conversation. He went on to teach at the Princeton University School of Architecture, at Rensselaer Polytechnic Institute, at Vassar, and as an adjunct at Columbia’s Graduate School of Architecture, Planning and Preservation, working at the edge of what he has described as context-aware fabrication — building processes in which a machine responds to the real, messy state of its materials rather than executing a fixed plan.
That line of inquiry led him to ETH Zurich, the Swiss institution that has become one of the world’s most important centres for construction robotics, where he completed his PhD in 2023. His doctoral work is the Rosetta Stone for everything Gravis is now doing commercially: he combined HEAP — an autonomous walking excavator — with planning algorithms to assemble large-scale walls out of irregular objects, including raw stone and demolition debris, dry-stacked without mortar.
Consider how hard that problem is. A dry-stone wall built from irregular rocks has no two pieces alike. There is no drawing that specifies where each stone goes, because the stones are whatever the site happens to provide. To stack them into a stable structure, the machine has to perceive each object’s actual geometry, reason about how it will sit against its neighbours, and plan a placement — over and over, adapting continuously. It is the opposite of factory automation, where everything is known in advance. It is autonomy in genuine, unstructured, real-world conditions: exactly the conditions of a construction site.
Why Gravis Retrofits Instead of Reinventing
That research background shapes the most important strategic decision Gravis has made: it does not build robots from scratch. It retrofits the machines contractors already own. The company’s hardware — the Gravis Rack — turns conventional excavators and loaders into systems that can be operated remotely, semi-autonomously, or fully autonomously, depending on the task and the operator’s comfort.
The logic flows directly from Johns’s intellectual roots. If the hard part of construction autonomy is the intelligence — perceiving an unstructured site and planning sensible action within it — then the machine underneath is a solved problem. Excavators are mature, reliable, and already in the hands of every contractor on earth. Reinventing the steel would be expensive, slow, and beside the point. The value Gravis adds is the mind, not the body. Retrofit is the fastest path to putting that mind on real sites at real scale.
It is also the most pragmatic path to adoption. A contractor does not have to buy a strange new machine, retrain crews on unfamiliar equipment, or bet a project on a startup’s hardware reliability. They keep the excavator they trust and add capability to it. That lowers the barrier to a first “yes” — and on a jobsite, the first yes is everything.
Copilot Before Autopilot
The product Gravis launched at CONEXPO reflects a second piece of hard-won pragmatism. Gravis Copilot is not full autonomy. It is a machine-guidance system that combines real-time terrain visualisation, excavation-depth guidance and human-form recognition with the operator still in control. It delivers a productivity gain today, in the operator’s hands, while laying the groundwork — through future software upgrades — for the autonomous capability to come.
This “copilot before autopilot” sequencing is a meaningful tell about how Johns thinks the technology will actually arrive. The fantasy version of construction robotics is an empty, self-driving machine. The realistic version is a human operator who becomes dramatically more effective, on a machine that is steadily taking on more of the work as the technology earns trust. Copilot lets a contractor capture value on day one rather than waiting for a future that is always eighteen months away. And every hour an operator runs it, the system is learning the site conditions that make fuller autonomy possible.
It is the same instinct visible in the way Gravis has gone to market: through partnership with the equipment establishment rather than in opposition to it. At CONEXPO, Hitachi Construction Machinery ran live demonstrations of one of its excavators executing repetitive tasks autonomously via a Gravis retrofit kit, and the company counts Develon among its OEM collaborators and Techint among its deployment partners, including projects in Argentina. Gravis is not trying to replace the machines; it is trying to upgrade them, with the blessing of the companies that build them.
The Bet on Physical AI
Gravis, founded in 2022, sits inside one of the most fashionable and contested categories in technology right now: “physical AI,” the application of machine intelligence to machines that act in the real world. The category has attracted enormous capital and equally enormous hype, and Johns is acutely positioned to know the difference between the two. His entire academic career was spent on the genuinely difficult problem at the heart of it — autonomy in unstructured environments — long before it became a venture-capital category.
That is why his CONEXPO framing is worth taking seriously rather than discounting as founder bravado. The claim that physical AI is moving “from the R&D lab to the American jobsite” is, in his case, almost autobiographical: he has personally carried this technology from an ETH research excavator stacking stones to a commercial retrofit winning a contractors’ vote in Las Vegas. The labour math underneath gives the bet its urgency — a large share of skilled machine operators are nearing retirement, and too few younger workers are entering the trade to replace them. Gravis is not selling autonomy as a way to remove people the industry has in abundance; it is selling it as a response to people the industry can no longer find.
Whether physical AI delivers on its promise across construction remains an open question, and the honest answer is that it will be settled jobsite by jobsite, not at trade shows. But if there is a founder whose entire path — architect, educator, roboticist — was a fifteen-year apprenticeship for exactly this moment, it is the one who learned to make machines think by teaching one to build a wall out of rocks.
