Google's Intrinsic open-sources its robot control stack
Two physical-AI stories landed on the same day, and together they draw the line the industry is converging on: the software layer is going free, and the sensing layer is going into the hands of one buyer.
Alphabet's Intrinsic open-sourced Intrinsic Core, the runtime, SDK and real-time control framework it had been keeping in-house — an Apache-2.0 layer that handles trajectory adjustment mid-movement, 6-DoF pose estimation, collision-avoidance motion planning, grasp planning, simulation and vision calibration, plus drivers for third-party grippers and 3D cameras. The pitch in the repository is unglamorous and correct: every robot project currently rebuilds the same foundations, so the unit that stewards the Robot Operating System has published them instead. Pose estimation is built on Nvidia's FoundationPose, which is the detail that tells you whose components end up inside a stack that calls itself hardware-agnostic.
Alongside it Intrinsic released an Open Machine Tending reference design for CNC machine tending, with modular support for hardware from Universal Robots and FANUC. The target is a specific, unglamorous market: tens of thousands of machine shops across the US and Europe fabricate intricate parts, and Intrinsic says only 8% of them use any automation at all, held back by cost and integration effort rather than by capability. If the foundation layer is free and the remaining work is configuration, the buyer stops being a systems integrator with a six-figure budget and becomes the shop owner who currently does the loading by hand. That is a much larger market and a much lower-margin sale — which is exactly why nobody had open-sourced their way into it.
Cognex agreed to acquire RealSense, the 3D depth-camera business Intel spun out in July 2025, for roughly $500 million in cash — with a further $56.5 million three-year retention program and about $50 million in restricted stock bringing the package to about $600 million. RealSense expects $80 million to $90 million of revenue in 2026, better than 50% growth on last year, and Cognex frames the purchase as entry to a robotic perception market it sizes at $600 million today and about $1.6 billion by 2030. The company's facial authentication line gets spun out into a separate business before closing, expected in the fourth quarter, subject to regulatory clearance.
The strategic read is straightforward: depth cameras are now a standard component on autonomous mobile robots, quadrupeds and humanoids, and Cognex — forty years in factory machine vision — would rather own the supplier than license from it. Set the two stories side by side and the split is clean. Intrinsic spent the day giving away the software that makes a robot work; Cognex spent $500 million buying the sensor that lets it see. When the control stack is freeware, the durable businesses sit in the silicon, the calibration data and the service contract.
We have tracked this deployment problem before — Anu ships six releases to cut a robot station deploy to days is the same argument from a smaller company.
What to watch: whether anyone outside Intrinsic ships a production cell on Intrinsic Core this quarter, and whether the Cognex deal clears regulators before the fourth quarter closes.
Does open-sourcing the robot stack actually open the market to small shops, or does the integration work stay expensive no matter who writes the runtime? Tell us in the comments.
Sources: Intrinsic Core (GitHub) · Open Machine Tending Solution (GitHub)
