Google's Intrinsic open-sources Intrinsic Core robotics platform under Apache 2.0

Industrial robotics has long been one of the least open corners of modern software. A manufacturer that wants a robot to pick a part off a conveyor typically buys a closed stack of controllers, drivers and programming tools, then pays systems integrators to make each cell work. That model has kept automation expensive and slow, especially for the small machine shops that make up the bulk of parts fabrication in the United States and Europe.

On September 22, 2026, that model got a significant jolt. Intrinsic, the intelligent robotics software company owned by Google parent Alphabet Inc., said at ROSCon 2026 in Toronto that it is open-sourcing the core components of its industrial robotics platform under a permissive Apache 2.0 license. The release is called Intrinsic Core, and it is available on GitHub at github.com/intrinsic-ai/intrinsic-core.

Intrinsic was founded in 2021 and was folded into Google proper in February 2026, when it stopped being an Alphabet Other Bet and became a distinct group inside the company. Chief executive Wendy Tan White reports to Hiroshi Lockheimer, and the unit plugs into Gemini models, Google Cloud and Google DeepMind. Intrinsic acquired Open Source Robotics Corporation, the commercial arm behind ROS, in December 2022, while ROS, Gazebo and ROSCon stayed with the nonprofit Open Source Robotics Foundation. ROS co-creator Brian Gerkey became Intrinsic's chief technology officer.

Forbes reported on September 22, 2026 that Intrinsic chief executive Wendy Tan White told CNBC that Sundar Pichai has called Intrinsic the Android of robotics. With the ROSCon announcement now on the record, that comparison started to look more literal than promotional.

Key Facts

SiliconANGLE reported on September 22, 2026 that Intrinsic, the intelligent robotics software company owned by Alphabet Inc., open-sourced the core of its industrial robotics platform under a permissive Apache 2.0 license at ROSCon 2026 in Toronto. The package is a preconfigured Robot Operating System compatible software environment designed to run on local hardware and assembled from reusable building blocks. Those blocks include Intrinsic Control, a hardware-agnostic real-time control framework that adapts a robot's trajectory mid-movement from sensor feedback; pose estimation built on NVIDIA's FoundationPose that delivers accurate 6-DoF part detection and removes the need for rigid fixtures; motion planning that auto-generates collision-avoidance paths instead of joint-by-joint programming; and grasp planning that lets grippers adapt automatically to an object's location and orientation.

Intrinsic wrote in its official blog post dated September 22, 2026 that these capabilities and services have huge potential for manipulation-based manufacturing and industrial applications, and that the simulation services inside Intrinsic Core are powered by Gazebo for testing, visualization and troubleshooting. Camera calibration automates both calibration and physical alignment, while Intrinsic-ROS drivers ship preconfigured for supported robots, third-party grippers and 3D cameras.

Alongside Intrinsic Core, the company released the Open Machine Tending Solution, an open reference design for AI-powered CNC machine tending, meaning robots that load and unload manufacturing equipment. The design supports hardware from Universal Robots A/S and FANUC Corp. out of the box, is customizable for foundation models such as NVIDIA FoundationPose, and runs on Intrinsic Core and the Open Robotics Suite.

The AI Insider reported on September 22, 2026 that developers can use the software for research projects or private production environments, and that applications built on Intrinsic Core work with other Intrinsic offerings, including Intrinsic Flowstate, advanced AI models and industrial grade cloud services, without refactoring or code rewrites. The publication added that Intrinsic plans additional open reference solutions for industrial automation applications.

The market context behind the release is stark. SiliconANGLE reported on September 22, 2026 that Intrinsic says the United States and Europe have tens of thousands of part-fabrication shops, yet only 8% use any form of automation, held back by cost and technical barriers. The same week, Intrinsic announced the winners of its first AI for Industry Challenge, focused on electronics assembly, which drew more than 5,000 developers and roboticists across 115 countries.

Analysis

Forbes reported on September 22, 2026 that the newly opened layer covers control, motion planning, grasp planning, simulation, pose estimation and camera calibration, all of it previously locked inside a proprietary product. What this really means is that Google is deliberately splitting industrial robotics into a commodity layer it gives away and a value layer it intends to charge for. Forbes framed the strategy as giving away the layer everyone needs while keeping the money-making parts, since the advanced AI models, Intrinsic Flowstate and the industrial grade cloud services remain proprietary.

The mechanics are explicit in Intrinsic's own positioning. Solutions built with Intrinsic Core automatically work with other Intrinsic offerings, including enterprise services with advanced AI models, Flowstate and cloud services, and need no refactoring or code rewrites. That is a textbook open-core funnel. The free tier removes the friction that keeps developers off the platform entirely, and the paid tier captures value once a robot cell moves from prototype into production.

The timing is not accidental. Google DeepMind shipped Gemini Robotics 2 on July 30, 2026, and Intrinsic now sits inside Google proper rather than outside it as an Alphabet Other Bet, plugged into Gemini models, Google Cloud and DeepMind research. Handing developers a free, ROS compatible control and planning stack gives those foundation models a far larger surface of real industrial hardware to run on.

There is history behind the bet, too. Forbes noted on September 22, 2026 that Intrinsic cut 20% of its workforce in January 2023, and that its December 2022 acquisition of Open Source Robotics Corporation gave it deep ties to the ROS community while the nonprofit Open Source Robotics Foundation kept ROS, Gazebo and ROSCon. Announcing an Apache 2.0 release on the ROSCon stage in Toronto spends that community credit at exactly the right moment.

Why It Matters

The number that matters most is 8%. If only about 8% of the tens of thousands of part-fabrication shops in the United States and Europe use any form of automation, the addressable market is not a niche, it is most of manufacturing. The two barriers Intrinsic names, cost and technical difficulty, are precisely what proprietary, integrator-heavy robot programming creates. A free, ROS compatible stack with hardware-agnostic control, automatic motion and grasp planning, and FoundationPose-based 6-DoF pose estimation attacks both at once.

It also redraws the competitive map. The Open Machine Tending Solution supports FANUC and Universal Robots hardware out of the box, which turns two of the biggest names in industrial arms into distribution channels for Intrinsic software rather than gatekeepers of it. NVIDIA, whose FoundationPose underpins the default pose estimation pipeline, is embedded in the core rather than sold as an add-on.

For working roboticists, the practical change is that the floor has been raised. Instead of writing drivers for every arm, gripper and camera, teams start from preconfigured Intrinsic-ROS drivers and a hardware-independent real-time control loop, then spend their time on the application itself. That is the same pattern that turned Android from a phone operating system into a platform, and it is the pattern Pichai was pointing at.

Next Up

Intrinsic said the Open Machine Tending Solution is the first of a series of open reference solutions, with more industrial automation designs to follow. The company has not said which tasks come next, but the roadmap logic is clear: each reference design pulls another manufacturing job onto Intrinsic Core and, by extension, closer to the enterprise services layer that carries the advanced AI models, Flowstate and cloud services.

What to watch is whether the hardware ecosystem follows. Intrinsic Core is already on GitHub under Apache 2.0, and the measure of success over the next year will be how many machine shops, integrators and robot makers build cells on it, and how many of those eventually pay for the layer above. If the Android comparison holds, the free part is the point, and the money arrives later.