19 August 2026

Image by Gravis Robotics

 

An excavator can weigh more than 100 tonnes and move enormous quantities of earth in a working day. The same raw power that makes heavy machinery indispensable to construction, quarrying and mining also makes mistakes potentially catastrophic.

In 2023, construction accounted for almost a quarter—23%—of fatal workplace accidents in the European Union. Workers can be struck by moving vehicles, trapped by equipment or exposed to unstable ground, falling material and poor visibility.

Gravis Robotics believes greater autonomy can reduce this human exposure while making machines more productive. The ETH Zurich spinout has secured $200 million (€172 million) from SoftBank in a Series A round that gives it a reported post-money valuation of $1 billion.

That makes the four-year-old Swiss company Europe’s latest robotics unicorn.

SoftBank is the sole investor in what Gravis describes as the largest Series A in construction robotics history. The funding follows a $23 million round—approximately €19 million at the time—completed in 2025.

Gravis’ technology addresses a particularly difficult form of physical AI. Many autonomous systems are designed primarily to recognise objects, avoid obstacles and navigate through an environment without changing it. An autonomous excavator must do almost the opposite: cut into the ground, move material and continuously adapt as its own actions transform the terrain.

The machine’s world is therefore not static. Every bucket movement changes the geometry, weight distribution and physical conditions of the next action.

Gravis is not attempting to manufacture an entirely new generation of autonomous construction equipment, however. Instead, it is building a layer of sensing, computing and robotic control that can be fitted to machinery contractors already own. Its proposition connects the raw physical power of industrial equipment with AI capable of observing and controlling how that equipment reshapes the physical world.

 

Why construction needs autonomous heavy machinery

 

Construction remains one of Europe’s most dangerous industries despite decades of improvements in equipment and workplace procedures. Autonomous machinery could remove operators from some hazardous areas, but its relevance extends beyond safety.

Contractors also face shortages of skilled workers, including experienced plant operators. The job requires an understanding of soil, machinery and site conditions that can take years to develop. An ageing workforce and difficulties attracting new recruits are making that knowledge increasingly scarce.

Gravis does not present its technology simply as an operator-replacement system. CEO and co-founder Ryan Luke Johns argues that experienced workers could instead supervise several autonomous machines.

“It’s not about taking operators out of the machine. It’s about getting machines to be 30 percent more productive, to get an operator to drive multiple highly productive machines.”

There is also an economic argument. Inconsistent digging can create rework, unnecessary fuel consumption and additional machine wear. Equipment can sit idle while contractors wait for an operator or surveyor. More precise and repeatable excavation could increase utilisation and help projects remain closer to schedule.

 

Physical AI that changes its environment

 

Founded in 2022, Gravis Robotics is led by CEO Ryan Luke Johns and co-founder and CTO Dominic Jud. ETH Zurich robotics professor Marco Hutter is a co-founder and board member.

The company’s central product is the Gravis Rack, a modular autonomous control kit mounted on an existing excavator or wheel loader. Gravis says it is compatible with machines ranging from approximately 10 tonnes to more than 100 tonnes.

The Rack uses cameras and lidar to scan the surrounding terrain, alongside positioning systems and onboard edge computing. Gravis says the system can continue operating without a permanent internet connection, which is important on remote or unfinished construction sites.

But perception is only the beginning. A self-driving vehicle generally tries to understand a changing world well enough to move safely through it. An autonomous excavator must understand how its own actions change that world.

Before digging, the system needs a model of the existing terrain and the intended result. It must then decide where to place the bucket, how deeply to cut and how to move the excavated material. After every action, it must scan the altered ground, compare the result with the plan and calculate the next movement.

This creates a continuous loop:

Perceive the terrain, act on it, measure the change and adjust.

Soil also behaves less predictably than the solid objects handled by a factory robot. Its response changes with composition, moisture, compaction and the angle at which force is applied. A bucket may collect more or less material than expected, while removing earth from one area can change the stability of another.

Experienced human operators manage these variations through sound, vibration, hydraulic resistance and years of accumulated judgement. Gravis must reproduce enough of that capability through cameras, lidar, positioning, machine telemetry and AI control policies.

The company says it trains its models in simulation, allowing machines to move enormous quantities of virtual earth before operating on a physical site. Simulation accelerates training, but it cannot reproduce every soil condition or equipment response. Closing the gap between virtual earth and wet, compacted or unstable ground remains one of the technology’s central challenges.

Vibration, mud, dust, rain and changing terrain also test the sensors and hardware. The system must translate its digital plan into precise hydraulic movements on machines produced by different manufacturers.

That is both the attraction and the engineering burden of retrofitting. Contractors avoid buying an entirely new fleet, but Gravis must make its control system understand how different machines interact with a physical environment that changes after every movement.

 

From operator assistance to autonomous excavation

 

Gravis offers a progression from operator assistance to autonomous operation.

Its system can provide three-dimensional guidance and hazard information to someone in the cab. It can also support remote operation or execute a defined excavation job autonomously. An operator can take manual control for more complicated work before returning repetitive tasks to the system.

“You press play, the machine will drive itself to the start and essentially do that entire job without intervention,” Johns says.

Remote operation depends on sufficiently fast and stable communication. Excessive latency would create a dangerous delay between an operator issuing a command, seeing the machine respond and reacting to a changing situation.

Full autonomy can reduce dependence on a continuous communications link because operational decisions are made onboard. Remote supervision and emergency intervention still require resilient communications and safe fallback behaviour. Gravis’ Slate tablet includes a wireless stop function intended to keep control available to the operator.

The long-term model is not necessarily a distant room filled with people continuously driving excavators over video. It is more likely to combine autonomous task execution with humans supervising machines and taking control when an exception occurs.

 

Autonomous excavation moves onto live construction sites

 

Gravis says its technology is being used across four continents, with customers and partners including Holcim, Taylor Woodrow and HD Hyundai. Applications include trenching, bulk excavation, truck loading, stockpile management and handling quarry material.

One of its most visible deployments involved Taylor Woodrow at Manchester Airport. Following earlier testing at a dedicated demonstration site, an autonomous excavator was used during construction of a new car park.

Taylor Woodrow described the project as the UK’s first large-scale use of autonomous excavation on an active construction site. That description comes from the contractor rather than an independently defined industry benchmark, but it moves Gravis beyond a controlled demonstration and into a working environment. 

Gravis is also leading an $8 million UK government-backed Connected and Automated Mobility Pathfinder project with Flannery Plant Hire. The programme is intended to retrofit six excavators for activities including trenching, bulk excavation and truck loading.

A separate rental partnership allows contractors to hire excavators already fitted with Gravis technology.

Rental could help the company move beyond demonstrations. Construction businesses may be reluctant to purchase an unfamiliar autonomy system before understanding the economics on their own sites. Hiring an equipped excavator allows them to test it on an individual project without making a permanent fleet decision.

 

Can Gravis deliver 30% productivity gains?

 

Gravis reports productivity improvements of up to 30% over manual operation on selected projects. Its website describes terrain-aware excavation capable of increasing throughput by that amount.

These reported productivity gains are especially significant because excavation efficiency is not determined only by how quickly a machine moves. It depends on how effectively each movement changes the terrain.

An autonomous system might improve output by selecting more consistent digging angles, filling the bucket more efficiently, reducing unnecessary repositioning and stopping closer to the required depth. Precision could also reduce over-excavation and the time spent correcting work.

The relevant measurement is therefore not simply cycles per hour. Contractors need to know how much usable work the machine completes, how much rework it avoids and how accurately the finished terrain matches the design.

 

A specialised corner of physical AI

 

Gravis is not alone in developing autonomous heavy machinery. US company Built Robotics offers an aftermarket autonomy system for excavators, with a particular focus on utility-scale solar construction. Teleo develops remote operation and supervised autonomy, while companies including SafeAI and Autonomous Solutions have targeted mining, quarrying and other industrial environments.

Large equipment manufacturers are also building more intelligence into their machines. Caterpillar, Komatsu, John Deere, Volvo and HD Hyundai have the advantage of controlling the underlying hardware, dealer networks and maintenance relationships.

Interesting in that regard is Gravis' vision for the future: a construction site where excavators, loaders and haulage equipment coordinate tasks with limited direct control. Its next challenge will be to prove that its feedback loop can work reliably across such mixed fleets, unpredictable terrain and active construction sites. 

 

 

 

Further reading on MoveTheNeedle.news:

Taking construction robots beyond the demonstration stage

From tractor driver to technology manager: how agricultural robotics is changing the job of farming

NEURA Robotics secures up to $1.4 billion as Europe intensifies its robotics ambitions