Alphabet's robotics subsidiary Intrinsic open-sourced its core robotics platform on September 22, 2026 at ROSCon 2026 in Toronto — releasing the foundational software stack for intelligent robots under the Apache 2.0 license on GitHub. The release is called Intrinsic Core, and it ships as a pre-configured, ROS-compatible environment that runs entirely on local hardware.
Intrinsic was spun out of Google's X lab in 2021 and had been developing its platform for industrial automation use cases. Until now, the core of that platform was proprietary. Opening it under Apache 2.0 — a permissive license with no copyleft restrictions — removes the licensing barrier for commercial use entirely.
What Intrinsic Core includes
The release bundles the essential layers for building sophisticated robotic applications:
- Motion planning: Collision-free path generation across both Cartesian tasks and C-space movements, with a unified API covering both domains.
- Perception: Standardized camera and point-cloud interfaces with out-of-the-box support for NVIDIA FoundationPose — a production-grade 6-DoF object pose estimator useful for pick-and-place in unstructured environments.
- Edge ML inference: On-device model serving with memory management, so inference runs on local hardware without cloud round-trips.
- Developer SDK: Base interfaces, serialization helpers, and data structures for building reusable robot skills and custom hardware assets.
- Kinematics library: Modeling and solver support for robotic manipulators.
- Hardware drivers: Device manifests and integrations for industrial peripherals.
Why ROS compatibility matters
ROS 2 (Robot Operating System) is the de facto middleware standard in both research and industrial robotics. By shipping a pre-built, hardware-agnostic environment that integrates natively with the ROS ecosystem, Intrinsic Core dramatically lowers setup cost for teams already using ROS 2. You get a tested, production-grade runtime to build on top of rather than assembling one from scratch.
The NVIDIA FoundationPose integration is also worth calling out specifically. Reliable 6-DoF pose estimation — knowing exactly where an object is and how it's oriented in 3D space — has historically required significant custom integration work. Having it ship as a standard, out-of-the-box capability removes a meaningful bottleneck for teams working on manipulation tasks.
The physical AI moment
Releasing this now, when both the hardware (capable edge compute, increasingly affordable manipulators) and the AI (vision-language models for robot policy) are maturing together, makes Intrinsic Core a timely foundation layer. The Apache 2.0 licensing means it can serve as the open-source core of commercial products without IP complications — positioning it as a potential shared infrastructure layer for the industry, similar to what ROS itself did for middleware.
For teams building robotic applications today, the barrier to getting started with a production-quality software stack just dropped significantly.