Gravis Robotics has raised $200 million from SoftBank in a Series A round intended to accelerate deployment of autonomous heavy construction machinery. The ETH Zurich spinout is developing retrofit control technology capable of operating excavators from several manufacturers, with systems already deployed on sites across four continents.
Construction machinery presents a different automation problem from factory robotics. An industrial robot normally works inside a controlled cell with known geometry and tightly specified parts, while an excavator repeatedly changes the environment in which it is operating. Soil composition varies, rocks and buried objects may be invisible, loads change through every bucket cycle, and the machine has to interpret hydraulic and mechanical feedback while maintaining safe movements around people, vehicles, and site infrastructure.
Gravis was founded in 2022 to address that problem using learning-based control, machine telemetry, and simulation. Instead of designing a proprietary excavator, it has built its system around the Gravis Rack, an autonomous control package intended to retrofit existing machinery. The company’s software has been installed on equipment from manufacturers including Caterpillar, Case, Develon, John Deere, JCB, Hitachi, Sumitomo, Yanmar, and Volvo.
The mixed-fleet approach removes one obvious commercial barrier. Contractors rarely operate machinery from a single manufacturer, and plant purchasing is influenced by dealer support, fleet history, application requirements, availability, and residual values. An autonomy system that required companies to discard usable equipment and adopt one dedicated robotic platform would place a large capital commitment in front of the productivity case it was supposed to prove.
Gravis instead describes its software as an operating layer that adapts to different machine characteristics. The company says the same core system can control equipment across substantially different size classes without being rewritten from scratch for every model. Training makes extensive use of simulated earthmoving before models are transferred into physical machines, reducing the amount of experimentation that has to occur on a live construction site.
The control stack covers assisted operation as well as autonomy. Gravis Copilot keeps an operator in the cab while adding three-dimensional guidance and hazard detection, while fully autonomous operation is intended to let a person supervise machines remotely rather than occupy every cab. Equipped excavators can also collect site information while they work, combining excavation with surveying and hazard-mapping functions.
Productivity is central to the commercial argument. Gravis says its systems have demonstrated gains of up to 30% against peak manual operation in suitable applications, although that remains a company-reported figure rather than a performance guarantee for every site. The harder test is whether those gains persist across changing ground conditions, attachments, machine sizes, project types, weather, and the ordinary variability of construction work.
The $200 million financing gives the business considerably more scope to run that test at scale. Gravis says the funding will support global deployment and recruitment as it installs autonomous systems across contractors’ equipment fleets. The company has operations in Zurich, Austin, and Oxford, putting it close to major European, North American, and UK construction markets.
Britain is part of that expansion. Gravis says it has been selected to lead an $8 million CAM Pathfinder programme with Flannery Plant Hire under which excavators will be retrofitted with the Gravis Rack. Flannery’s large plant fleet provides an environment in which machines move between customers, operators, and projects rather than remaining permanently attached to one controlled demonstration site.
That should expose the integration issues that determine whether construction autonomy becomes an operating tool rather than an occasional showcase. Retrofit hardware has to coexist with the original machine’s hydraulics, controls, diagnostics, safety systems, and maintenance requirements. Contractors also need practical arrangements around connectivity, remote supervision, cyber security, mobilisation, breakdown response, site planning, and the handover between manual and autonomous modes.
The system has to be maintainable by an organisation whose primary job is construction, not robotics research. A productivity gain can disappear quickly if commissioning takes too long, specialist support has to attend every move between sites, or a fault leaves an expensive machine standing while conventional equipment continues working around it. Standardised installation and support may therefore be as important to adoption as the underlying AI models.
Gravis has chosen to tackle that problem without asking contractors to replace their mixed fleets. Assisted and autonomous modes also allow deployment to happen in stages, giving operators a route to use guidance and hazard detection before progressing towards remote supervision of repetitive earthmoving tasks.
SoftBank’s investment gives the company the resources to move beyond isolated demonstrations, but the next useful metric will be operational rather than financial. Gravis now has to show that machines remain productive after the commissioning engineers have gone home, across enough sites and brands for autonomy to become part of ordinary plant utilisation rather than a special project.
