Google's Intrinsic Open-Sources Its Industrial Robot Software

Intrinsic, the robotics unit Alphabet folded into Google in February 2026, released Intrinsic Core on Sept. 22 at ROSCon in Toronto, open-sourcing the real-time control, motion planning and pose estimation software it uses in its own industrial deployments under a permissive Apache 2.0 license.

The release, tagged 20260922.0 on GitHub at intrinsic-ai/intrinsic-core, runs on Ubuntu 24.04 or 26.04 and the ROS 2 Lyrical distribution, according to Unite.AI's account of the launch. It is the same software Intrinsic says it runs day to day in manufacturing deployments, not a stripped-down demo version.

What ships free: control, grasping, a digital twin and one reference design

The package is called Intrinsic Control, internally abbreviated ICON, a hardware-agnostic real-time control framework that adapts to sensor input rather than following a fixed script. Around it sit motion planning that generates collision-free paths automatically, grasp planning that adjusts gripper position using live sensor feedback, camera calibration, a native digital twin, and Intrinsic-ROS drivers for supported hardware. Pose estimation runs on NVIDIA's FoundationPose model for six-degrees-of-freedom accuracy. Intrinsic also shipped a reference design called the Open Machine Tending Solution, aimed at getting a machine shop's parts-loading robot running without custom integration work, built atop hardware from partners including Universal Robots, which shipped its own Gen 7 cobot line without disclosing pricing.

Component	What it does
Intrinsic Control (ICON)	Real-time, hardware-agnostic control loop
Motion and grasp planning	Collision-free paths, sensor-adjusted grasping
Pose estimation	NVIDIA FoundationPose, 6 degrees of freedom
Open Machine Tending Solution	Reference design for machine-shop part loading
Supported grippers and sensors	FANUC, Universal Robots, ATI, Schunk, Robotiq, Basler

Who it works with, and what Intrinsic doesn't say

Intrinsic names seven hardware partners: robot makers FANUC and Universal Robots, and gripper, sensor and camera vendors ATI, Schunk, Robotiq and Basler, alongside NVIDIA for perception. That list excludes Boston Dynamics, KUKA and ABB, three of the largest industrial arm makers, and neither Intrinsic's post nor Unite.AI's coverage says whether integrations for them are planned or simply unsupported. Intrinsic also does not name a single open source competitor, despite releasing into a field that already includes ROS-Industrial and, as of the same week, NVIDIA's own Isaac ROS agent tooling.

The Red Hat model its CEO promised four years earlier

Intrinsic Chief Executive Wendy Tan White told IEEE Spectrum in December 2022, when Intrinsic first partnered with Open Robotics, that "our whole mission at Intrinsic is about democratizing robotics. The demand is there now, but access is still limited," and that Intrinsic would act as "the Red Hat to the Linux of OSRF," the foundation that stewards ROS itself. Intrinsic Core is the clearest test yet of that promise: a Red Hat model means charging for support, integration and hosted services around software that stays free, rather than gating the software behind a license fee, and the Apache 2.0 terms on Intrinsic Core are consistent with that structure four years later.

The 8% figure nobody outside Intrinsic has checked

Intrinsic frames the Open Machine Tending Solution around a statistic that only 8% of machine shops use automation today, the justification for why a pre-built reference design matters more than raw software capability. Neither Intrinsic's post nor the coverage of it names the survey or the year behind that number, and no independent industry body is cited confirming it. It is the kind of figure that is plausible on its face, machine shops are a famously under-automated corner of manufacturing, but it is also exactly the sort of vendor-supplied statistic that should carry a source and does not.

The same week, Intrinsic announced the winners of its first AI for Industry Challenge, an electronics-assembly competition that drew more than 5,000 participants across 115 countries, according to Unite.AI. Neither the winning teams nor their results were named in the coverage available, which makes the 5,000-participant figure a measure of entry interest rather than of anyone having solved anything.

What incorporation into Google changed

Intrinsic launched in 2021 as an Alphabet company under its Other Bets structure, the same arrangement that keeps Waymo and Verily operating apart from Google's core business. Alphabet folded Intrinsic into Google in February 2026, a move that typically means closer access to Google's cloud, sales and legal infrastructure in exchange for less operating independence. An open source release under a Google-run legal team is a different proposition than one from a standalone Other Bet: Apache 2.0 contributions now sit inside Google's existing patent grant and contributor framework, which is a more predictable structure for outside developers to build on than a smaller company's bespoke terms would have been, even if neither Intrinsic's post nor Unite.AI's account frames the move that way.

What would change this read

The test of the Red Hat comparison is not the license text but the paid tier: whether Intrinsic starts selling support contracts, hosted deployment or certified integrations on top of the free Intrinsic Core stack, the way Red Hat built a business on free Linux. Until Intrinsic publishes pricing for anything beyond the open packages, or names its Open Machine Tending Solution customers the way OpenArm named a per-pair price for its open source robot arm, the honest description of Intrinsic Core is a credible, well-documented release of real production code with an unverified adoption statistic attached to it, not yet a business model.