Intrinsic, Alphabet's industrial robotics software company, released Intrinsic Core on September 22, a set of open components for building and running robotics applications from simulation to hardware. The release is not a robot or a new general-purpose artificial intelligence model: it is a software foundation combining local execution, real-time control, perception, planning, and development tools.

An open, ROS-compatible foundation

The repository is available on GitHub under the Apache 2.0 license. Intrinsic presents it as a way to keep every team from having to integrate the foundational layers of a robotic workcell from scratch. The system uses ROS 2 and offers interoperability with the ROS ecosystem, along with an environment configured to connect services that are normally assembled separately.

The released modules include Intrinsic Control, an engine for coordinating motion and hardware in real time; kinematics and planning libraries; camera and point-cloud interfaces; a local inference environment; a development kit; API definitions; and an abstraction layer for industrial devices. That separation allows arms, grippers, or sensors to be changed without rewriting the entire application, although actual compatibility will depend on the drivers and assets available for each piece of equipment.

The perception layer includes integration with Nvidia's FoundationPose to estimate the position and orientation of parts in six degrees of freedom. Intrinsic also incorporates a digital twin and simulation with Gazebo. The combination is intended to let a developer test a solution virtually and then move it to hardware using the same interfaces, reducing the integration work between perception, planning, and control.

A reference use case for factories

Alongside the core, the company released Open Machine Tending Solution, a reference implementation for tending CNC machines. It includes assets, skills, and configurations that can be adapted to an actual workcell. Its value lies not in automatically solving any manufacturing process, but in showing how the project's components connect in a concrete industrial application.

The repository confirms that Intrinsic Core includes a local environment, SDK, and vendor-independent control framework. It also clarifies that the project is not an officially supported Google product. That distinction matters: open source does not mean a factory receives guarantees of integration, functional safety, or maintenance for every configuration. Companies will still have to validate response times, equipment compatibility, stopping mechanisms, access controls, and behavior in the event of failures before using it in production.

Intrinsic maintains that solutions built with Core can move to its enterprise services and Flowstate without rewriting. That path offers continuity between the open layer and the commercial platform, but it also makes the architecture an entry point into the company's ecosystem. Developers will need to evaluate which components remain portable and which depend on Intrinsic-specific services.

The release matters because it makes publicly available components that normally remain inside proprietary automation platforms. For research projects, integrators, and manufacturers, having control, perception, simulation, and a production reference in a single repository can reduce the initial friction. Even so, the decisive test will be external: adoption by the ROS community, third-party contributions, support for more hardware, and verifiable results in actual plants.

Intrinsic Core arrives as infrastructure, not as a promise of universal autonomy. Its immediate contribution is to make a larger portion of the stack connecting perception and planning models to physical machines inspectable and modifiable. In industrial robotics, where a software error can turn into motion, that openness also facilitates auditing; it does not replace safety engineering.

Sources: Intrinsic announcement, product page, and official repository.

— Lía Torres, AI News