Construction robots are starting to roll out like software, and Caterpillar knows the drill

Caterpillar’s autonomy story has often sounded like a hardware story: giant trucks, big engines, and sprawling sites. This week, the emphasis changed. Caterpillar’s message was that getting systems deployed at scale is the hard part, and that its experience comes from years of operating autonomous machines in remote locations where support is limited.

At the Ai4 conference in Las Vegas, Caterpillar CTO Jaime Mineart said the company is bringing what it learned from mining automation to AI deployments in “more dynamic environments,” including construction sites, quarries, and jobsites, according to TechCrunch.

The timing matters because construction autonomy is moving past demonstrations and into buying decisions. Robotics 24/7 reported that Switzerland-based Gravis Robotics raised a $200 million Series A from SoftBank. Gravis said it is the largest Series A in construction robotics history, and said the money will be used to scale autonomous heavy machinery globally. That level of funding can move discussions from an “R&D roadmap” to “fleet rollout expectations” for owners, EPCs, and the contractors running the equipment.

The next constraint is repeatable rollout, not one-off autonomy.

TechCrunch described Caterpillar’s advantage as operational experience. The company has already sold automated haul trucks, drills, underground loaders, dozers, and remote-controlled construction equipment, and it supports those products with a software command center, fleet management, and terrain intelligence tools. The theme is that autonomy is not limited to a single model on a single machine. It behaves more like a fleet layer that needs monitoring, updates, and predictable performance across sites.

Gravis is tackling a similar rollout problem from the software side and then applying that software to heavy equipment. Robotics 24/7 reported that the company was founded in 2022 and spun out of ETH Zurich, and that it is pursuing “software-defined intelligence” in a sector long oriented around mechanical hardware. Gravis says it trains models in simulation to address the sim-to-real gap, and it says it wants performance that can transfer across manufacturers rather than being tied to one machine type.

For operators, the change also shifts who owns the work internally. A machine-first autonomy pilot often sits with equipment managers and project leadership. A fleet-first rollout can quickly turn into an IT/OT program, spanning edge compute, networking, identity and access, data retention, and rules for what gets updated and when.

Connectivity is turning into a budget line in autonomy programs.

A system is only as “autonomous” as its ability to keep operating safely when a site is remote, the network is congested, or the backhaul goes down.

SpaceNews reported that Starlink’s government offering, Starshield, has received more than $300 million in task orders under the U.S. Space Force’s Proliferated Low Earth Orbit (PLEO) program from May 2024 through February 2025. Over the same period, SpaceNews put OneWeb at about $7.5 million. Those numbers come from defense procurement, but they show that low-latency, redundant LEO communications are being purchased at scale for distributed operations.

SpaceNews also cited Government Accountability Office estimates that, in fiscal year 2023, the U.S. Department of Defense spent $788 million on commercial satellite communications services, with 20.7% going to LEO satcom and 74.8% to geostationary services. The point for autonomy teams is not to mirror defense programs. It is that connectivity design, hybrid networking, and terminal logistics can be treated as deliberate design choices rather than afterthoughts.

Standards pressure rises as machines become software platforms.

As autonomy becomes deployable software, shared terms and standards can matter as much as vendor claims. According to the SAE International portal, the organization organizes information on standards activity and professional development across mobility domains.

That becomes a practical issue for enterprises using multiple contractors across multiple projects. When autonomy is bought as a capability, the project owner is often the party that has to require interoperability, define data interfaces, and insist on consistent validation practices, especially when different manufacturers and autonomy suppliers are working on the same site.

How this shows up in specs being written now.

The funding and the messaging point in a similar direction. TechCrunch reported that Caterpillar is positioning deployment expertise as a differentiator. Robotics 24/7 reported that Gravis has the capital to move faster and expand more broadly. SpaceNews reported that defense buyers are already purchasing LEO connectivity as an operational asset. Together, it suggests that construction and earthmoving autonomy evaluations are tightening into a checklist: remote operations maturity, connectivity plans, and fleet management that looks like software.

• For fleet and project owners: In autonomy RFPs, require a clear operating model for the “command center,” including who monitors, how exceptions are handled, and what happens when communications drop. Caterpillar’s description of command center and fleet tools (TechCrunch) is a useful reference for what to ask vendors to document.
• For IT/OT teams: Treat backhaul as a dependency with measurable performance targets. SpaceNews’ PLEO task-order figures and DoD satcom mix (GAO estimates cited by SpaceNews) are a concrete reminder that hybrid comms architectures are normal in high-consequence operations, and they may be appropriate on remote sites.
• For procurement: Ask autonomy suppliers to specify their sim-to-real approach, update cadence, and cross-OEM generalizability. Gravis’ stated focus on simulation and manufacturer-spanning performance characteristics (Robotics 24/7) gives a template for the kind of technical commitments that can be written into SOWs.