Alphabet's Intrinsic open-sources its industrial robotics stack under Apache 2.0

Alphabet's robotics subsidiary Intrinsic announced Intrinsic Core at ROSCon 2026 in Toronto this week, releasing the foundational infrastructure it uses for industrial robot deployments as open-source software under Apache 2.0. The move hands developers the same perception, planning, and real-time control stack that Intrinsic uses in live manufacturing environments — not a stripped-down demo version.

What's in the release

Intrinsic Core ships as a pre-configured ROS 2-compatible environment that runs on local hardware. The key components include:

Real-time control framework — adjusts a robot's path mid-move from live sensor feedback, compatible across hardware platforms
Pose estimation — built on NVIDIA FoundationPose, allowing robots to locate parts without rigid fixtures or precise pre-positioning
Motion planning — generates collision-free paths across Cartesian tasks and wide-envelope C-space movements through a unified API
Grasp planning — adapts gripper behaviour to how an object is actually sitting, reducing fixturing requirements
On-edge ML inference engine — serves models locally with memory management and inference scheduling
Simulation, calibration, and ROS drivers — hardware device drivers and manifests for industrial peripherals

The goal is to let teams compose robotic behaviours from pre-built blocks rather than coding perception, planning, and control from scratch — a problem that has historically made industrial robotics projects expensive and slow to prototype.

Why this release matters

Industrial robotics has long suffered from a fragmented software ecosystem: ROS provides a foundation but leaves large gaps in control, perception integration, and hardware abstraction. Commercial alternatives lock customers into proprietary tooling. Intrinsic Core positions itself as a middle path — open and composable, but production-hardened from real manufacturing use.

Intrinsic CTO Brian Gerkey framed this as a durable commitment rather than a one-off open-source drop, with the company committing to ongoing maintenance and community governance. That addresses the biggest concern developers have about corporate open-source releases: abandonment after the PR cycle ends.

The timing is strategic. Physical AI — robots that learn from sensor data and adapt to unstructured environments — is the industry's next major frontier after language models. By open-sourcing the foundational layer, Intrinsic is positioning itself as the infrastructure layer for that shift, the way AWS became infrastructure for cloud applications.

Availability

Intrinsic Core is available on GitHub under Apache 2.0. The release includes full documentation, hardware driver manifests for major industrial peripherals, and the same base interfaces Intrinsic uses internally. Developers already on ROS 2 can integrate it without a full platform migration.