intrinsic-ai/intrinsic-core

What it solves

Intrinsic Core simplifies the development of AI-enabled industrial robotics applications. It removes the need for developers to manually build low-level system plumbing, write custom perception wrappers, or rewrite drivers when switching robot hardware, providing a standardized path from prototype to deployment.

How it works

It provides a local runtime and SDK that integrates several core robotics modules:

• Runtime & Control: A containerized execution engine (k3s) manages process lifecycles, while a real-time motion engine (ICON) handles kinematics and hardware coordination via a Hardware Abstraction Layer (HAL).
• Planning & Perception: A motion planning engine generates collision-free paths, and a perception layer provides 6-DoF pose estimation for 3D parts using NVIDIA FoundationPose®.
• Inference: A dedicated local ML execution engine handles model serving and memory management at the edge to guide robot manipulation.
• Interoperability: It maintains seamless compatibility with ROS 2, bridging modern C++/Bazel development with the broader robotics ecosystem.

Who it’s for

It is designed for robotics developers and engineers building industrial applications, such as machine tending, who need a hardware-agnostic framework to deploy AI-driven perception and control.

Highlights

• Hardware Agnostic: Use a unified abstraction layer to swap robot arms and grippers without rewriting drivers.
• Integrated AI Perception: Out-of-the-box support for high-accuracy 6-DoF pose estimation.
• Real-time Execution: Deterministic control loops capable of switching controllers within a single cycle based on sensor feedback.
• Reference Solution: Includes an Open Machine Tending Solution (OMTS) as a functional template for real-world manufacturing use cases.