Intrinsic Gives Away the Industrial Robot Stack

Intrinsic, the Alphabet robotics company that moved into Google this year, has open-sourced the foundation of its industrial robotics platform. Intrinsic Core launched September 22 at ROSCon 2026 in Toronto, published on GitHub under the permissive Apache 2.0 license.

This is not a demo repo. It is the plumbing Intrinsic has spent five years building to make robot arms programmable by people who are not specialist robot integrators:

• Intrinsic Control, a hardware-agnostic real-time control framework that can adapt trajectories on the fly from sensor input
• Pose estimation built on NVIDIA's FoundationPose, for locating parts in six degrees of freedom
• Motion planning that generates collision-free paths instead of hand-taught joint positions
• Grasp planning that adjusts the gripper to where an object actually is and how it is oriented
• Simulation (powered by Gazebo) and camera calibration services
• ROS drivers for supported robots, grippers and 3D cameras

The repository packages this as a local runtime, SDK and containerized execution engine targeting Ubuntu and the current ROS 2 distribution. It runs on your own hardware, not in Google's cloud.

Alongside it, Intrinsic released the Open Machine Tending Solution, a reference design for AI-enabled CNC machine tending — loading and unloading parts from machine tools — supporting FANUC and Universal Robots hardware.

Why the timing is not accidental

The software industry learned long ago that the layer everyone needs but nobody differentiates on is the layer you give away. Industrial robotics has been slow to apply the lesson. Each arm vendor ships its own controller, programming environment and teach pendant; integrators rebuild the same perception-and-planning glue for every cell. That fragmentation is a big reason automation has stayed concentrated in large plants that can afford bespoke integration.

Intrinsic's bet is that the fragmentation is the bottleneck, not the hardware. By releasing control, perception and planning as a common, ROS-compatible substrate, it lowers the cost of building a cell to roughly the cost of configuring one.

It also clears the way for the part Intrinsic keeps. The company's blog says solutions built on Core automatically work with its enterprise offerings, including advanced AI models and Flowstate, its commercial development environment. That is the Android pattern, and Forbes reached for exactly that comparison: make the base layer free and universal, charge for the services that sit on top, and let the ecosystem do your distribution.

Why ROSCon, and why Gerkey matters

Launching at ROSCon, rather than a Google event, is the tell. The Robot Operating System is the de facto open framework in robotics research and a growing share of commercial work, and its community is allergic to vendors that throw code over the wall and leave.

Intrinsic has unusual credibility here. Its CTO, Brian Gerkey, co-founded Open Robotics, the organization behind ROS, and chairs its board; Intrinsic acquired Open Robotics' commercial arm in 2022. Gerkey told The Robot Report the release is "not us coming by and dropping off some code," describing it as a durable commitment. Geoffrey Biggs, CTO of the Open Source Robotics Foundation, called it a first step and a demonstration of where robotics development is heading.

Durability is the right thing to promise, because it is the thing developers will test. A permissive license makes adoption easy; only sustained maintenance makes it safe. Contributions require a Google Contributor License Agreement, which keeps Google firmly in control of the upstream.

Why machine tending is the wedge

The reference design is the most commercially pointed part of the release. CNC machine tending is repetitive, well-defined work sitting inside thousands of small and mid-sized fabrication shops, most of which have never automated because the integration bill does not pencil out for a handful of machines. SiliconANGLE, citing Intrinsic, puts the share of U.S. and European machine shops without automation at 92%.

That is the market the open stack is designed to unlock. If a shop — or a small integrator — can take a working reference cell on FANUC or Universal Robots hardware, point a camera at the fixture and let pose estimation and grasp planning handle part variation, the economics of automating a few machines change.

The limits

Scope is narrow for now. Coverage so far describes Intrinsic Core as focused on robot arms and commercial manipulation — not mobile robots or humanoids — and best suited to developers who already know ROS. The "physical AI" framing is aspirational; what ships today is a well-engineered classical robotics stack with learned perception at key points, not a general-purpose robot brain.

That is arguably its strength. Foundation models for robots still need something to execute their decisions reliably in real time, calibrated to real cameras, planning around real obstacles. That execution layer is exactly what Intrinsic has open-sourced.

Why it matters

Industrial robotics is about to have an open default it did not have before, from a company with Google's resources behind it and ROS's founder at the technical helm. Arm makers that have relied on proprietary software lock-in now compete against a free, hardware-agnostic alternative their customers can adopt without asking permission.

The open question is whether the community treats Intrinsic Core as shared infrastructure or as a Google on-ramp to Flowstate. Both can be true, as they are with Android. What matters is that the cost of building a capable robot cell just dropped, and the pressure on everyone still charging for that layer went up.