Google's Intrinsic Open-Sources Its Robotics Core

Google's robotics unit just gave away a piece of infrastructure it spent years building for itself. At ROSCon 2026 in Toronto on Tuesday, Intrinsic released Intrinsic Core, the real-time control, perception and planning software behind its own industrial robots, under the permissive Apache 2.0 license. Forbes called it "the Android of robotics," a comparison that only works because of who is doing the releasing. Intrinsic owns Open Robotics, the nonprofit-turned-Alphabet-subsidiary behind ROS, the operating layer most of the robotics world already runs on. Giving away the layer built on top of that is a different kind of move than a startup open-sourcing a side project.

Table of contents

• What Intrinsic Core actually contains
• A reference design aimed at a real gap
• Why Intrinsic controlling ROS made people nervous
• The strings attached
• What this means for robotics developers
• The open question
• Frequently asked questions

What Intrinsic Core actually contains

Intrinsic's own announcement and The Robot Report's coverage describe five pieces bundled into one package. A hardware-agnostic real-time control framework, which Intrinsic calls ICON, can switch controllers mid-cycle using live sensor feedback rather than following a fixed script. Pose estimation is built on Nvidia's FoundationPose, letting a robot locate a part without a rigid fixture holding it in place. Motion planning generates collision-free paths, and grasp planning adapts a gripper's approach to how an object happens to be sitting. Simulation, calibration services and ROS-compatible drivers round out the stack.

AiCybr's technical read of the GitHub repository, created September 8 and actively updated through the announcement, notes the code is split into modules with distinct deployment roles rather than shipped as one monolithic build. That matters for teams who want part of the stack, say the control layer, without adopting all of it.

Brian Gerkey, Intrinsic's CTO and the original co-creator of ROS, framed the release around a gap-filling logic in comments to The Robot Report: the team looked at its own software and asked which parts filled holes the existing ROS ecosystem did not already cover well, rather than releasing everything it had built.

A reference design aimed at a real gap

Alongside Intrinsic Core, the company released the Open Machine Tending Solution, a reference design for AI-enabled CNC machine tending that runs out of the box on hardware from Universal Robots and FANUC. Shopifreaks reports Intrinsic's own figure for why this matters: only about 8% of machine shops in the US and Europe use any automation at all.

That number is the real target of the whole release. Machine tending, loading and unloading parts from a CNC machine, is repetitive, physically demanding and chronically understaffed work. A small shop that wants to automate it today has to stitch together a robot arm, a vision system, motion planning and safety logic from scratch, usually with contractors who charge accordingly. A working reference design that already talks to common arms lowers that barrier in a way a raw software library alone would not.

Why Intrinsic controlling ROS made people nervous

The backstory is what gives this release its edge. Intrinsic acquired Open Source Robotics Corporation, the commercial arm behind ROS, in December 2022. Forbes' account of that period is blunt: the robotics community was nervous about what an Alphabet subsidiary might do with that much influence over infrastructure the whole industry depends on. Gerkey's response at the time, as Forbes recounts it, was to ask people to judge the company by its actions rather than its ownership structure.

Nearly four years later, Intrinsic Core is one answer to that challenge, though not a complete one. Releasing widely used internal infrastructure under Apache 2.0, a license that permits commercial use, modification and redistribution with minimal restriction, is a stronger signal than a blog post promising good intentions. It does not retroactively settle every worry the acquisition raised, but it is a concrete data point in the community's favor rather than a rhetorical one.

The strings attached

Two caveats are worth knowing before treating this as a finished product. The GitHub repository itself states that Intrinsic Core is an Intrinsic project, not an officially supported Google product, and AiCybr's write-up flags that production evaluations should independently validate hardware compatibility, real-time behavior and safety architecture against the target system before deployment.

Intrinsic's own blog post is candid about where the project stands: using Intrinsic Core still requires basic proficiency in robotics, and the company describes this as the first of several steps in an "early days" open-source effort rather than a finished platform. It is also worth noting, per TheAIInsider's coverage, that applications built on Intrinsic Core can connect to Intrinsic's paid products, Flowstate, AI models and cloud services, without rewriting code. The open core sits next to a commercial layer, which is a common and reasonable structure, but not a purely philanthropic one.

What this means for robotics developers

For a team already using ROS 2, the practical pitch is straightforward: pick up real-time control, pose estimation and motion planning that Intrinsic already runs in production, instead of building or licensing each piece separately. AiCybr describes the value proposition as breadth spanning perception, planning, control and hardware abstraction in one interoperable package, which is unusual for a single open-source release in this space.

For a small manufacturer eyeing automation for the first time, the Open Machine Tending Solution is the more direct entry point, since it is meant to run with less custom integration work than the core libraries alone would require. Neither path removes the need for the hardware itself or the safety engineering around a robot working near people, and Intrinsic's own disclaimer language makes that responsibility explicit rather than implied.

The open question

The release lands during a broader industry push toward what Intrinsic and others call physical AI, extending large-model capabilities from text and images into robots that act in the physical world. Forbes notes the connection to Google DeepMind's Gemini Robotics 2 work remains loosely defined for now, and neither Intrinsic's announcement nor the reporting reviewed here spells out how the two efforts will combine.

What is clear is narrower and more concrete: a company that controls the dominant open robotics framework has given away a meaningful layer built on top of it, with real usage behind the code and real limits stated alongside it. Whether that becomes the default toolkit for industrial robotics, or one option among several serious ones, will depend on adoption over the next year, not on the announcement itself.

Frequently asked questions