Google Just Open-Sourced the Robot Brain That Runs Its Core Robot Software

On September 22 at ROSCon in Toronto, Intrinsic, Alphabet's industrial robotics company, released Intrinsic Core under the Apache 2.0 license. Real-time control, motion planning, grasp planning, pose estimation, and simulation are all open source and free to use commercially.

Most industrial robot software is still sold as closed, vendor-specific platforms. Intrinsic just made the foundational layer available to anyone who wants to build on it.

What Intrinsic Core actually includes

The release has six pieces. Intrinsic Control is a real-time control framework that works across hardware brands. Pose estimation built on NVIDIA's FoundationPose lets a robot locate objects from a CAD model or a few reference images, without training a custom model for each part. Motion planning generates collision-free paths. Grasp planning handles pick-and-place. Gazebo-powered simulation lets you test a cell before you touch hardware. Camera calibration and ROS drivers round it out.

Intrinsic also released the Open Machine Tending Solution, a reference design for loading and unloading CNC machines with FANUC and Universal Robots arms. For machine shops, that's the most practical part of the announcement: a working starting point, not just a toolkit.

Everything is ROS-compatible. ROS is the most widely used open-source robotics framework, so these tools fit into setups many robotics engineers already know.

One scope note: Core targets robotic arms. It doesn't cover mobile robots or humanoids.

What stays paid

Google doesn't open-source things out of generosity. It does it when the real value sits at a different layer. With TensorFlow, that layer was cloud compute. With Android, it was search and ads.

Intrinsic is explicit about where its paid layer is. Flowstate (its development platform), Intrinsic Intelligence, the Intrinsic Vision Model, and its enterprise cloud services stay proprietary, and anything built on Core works with those paid offerings. That's the play: give away the foundation, and sell the tools and AI that make it production-grade.

Why this matters beyond the factory floor

The software barrier to physical automation just dropped. You still need hardware, integration, and someone who knows ROS, and those remain the real costs. But teams no longer have to license or build a control, planning, and vision stack from scratch before they can prototype a robotic cell.

Expect integrators and startups to start packaging Intrinsic Core into ready-made solutions for specific jobs like machine tending, packing, kitting, and inspection. The first ones to show working deployments will set the pricing for everyone else.

What to do with this

Watch the Intrinsic Core GitHub repo (github.com/intrinsic-ai/intrinsic-core). Documentation quality and community activity over the next few months will tell you how production-ready it is.

If you have a client with repetitive tasks at a fixed station, like tending a CNC machine, packing, sorting, or inspecting parts, this is now a realistic conversation. Start with the machine tending reference design if the fit is there.

And if you're building an AI consulting practice, put physical AI on your radar. The gap between "AI for screens" and "AI for the physical world" is starting to close.

The bottom line

Google just gave away the foundation of an industrial robot software stack and kept the premium layer for itself. For small and mid-size manufacturers, the cost of experimenting with robotic arms dropped a lot. For consultants, it's an early look at where the next wave of automation work is headed.