Alphabet's Intrinsic open-sources core robotics platform under Apache 2.0

Intrinsic was founded in 2021 and was incorporated into Google's operations in February to accelerate physical AI, a field in which sensors, actuators and control systems let machines perceive, reason, act and learn in real-world environments. Unlike pure software, physical AI bridges bits and atoms and can allow robots, self-driving cars, drones and smart infrastructure to take action.

Progress in physical AI has been stalled by the complexity of developing underlying software and training AI models for intelligent robots, according to the company. Google and Intrinsic believe a reusable foundational software layer can speed development so that engineers do not have to rebuild everything for each new robot.

Intrinsic Core is a preconfigured, Robot Operating System-compatible software environment designed to run on local hardware. It provides reusable building blocks for industrial robotics systems. Among them is Intrinsic Control, a hardware-agnostic real-time control framework that can dynamically adapt a robot's trajectory mid-movement based on sensor feedback. Intrinsic said this makes it easier to use arms, grippers and other hardware without rewriting underlying software and drivers each time.

The platform also includes Pose Estimation, based on Nvidia Corp.'s FoundationPose, which lets robots dynamically detect and handle parts and removes the need for rigid, expensive physical fixtures. A motion planning capability automates collision-avoidance path generation, eliminating manual joint-by-joint position-controlled programming. A grasp planning function lets different robotic grippers automatically and dynamically adapt to the location and orientation of objects they are trying to grasp.

Intrinsic Core provides simulation and calibration services that allow users to simulate, visualize and troubleshoot robotics and calibrate vision systems through a single interface. Intrinsic-ROS drivers make it easier to integrate sensors with third-party grippers, 3D cameras and other hardware components.

To show how the platform can help developers, Intrinsic released the Open Machine Tending Solution, an open reference design for AI-powered CNC machine tending tasks. It supports modular customization for different hardware setups from major robotics manufacturers, including Universal Robots A/S and FANUC Corp. Intrinsic said the design is meant to accelerate automation in the machine building industry.

The company said the U.S. and Europe have tens of thousands of shops that fabricate intricate parts for machines across industries, but only 8% use any form of automation because of high costs and technical barriers. By offering a customizable starting point, Intrinsic believes its reference design can help smaller fabricators automate lower-value tasks and accelerate automation.

More broadly, Intrinsic said it hopes to foster a larger robotics developer ecosystem based on shared open-source tools and to accelerate adoption of physical AI across industries.