NVIDIA ships Isaac ROS 5.0, adding AI agent workflows to open-source robotics development

NVIDIA released Isaac ROS 5.0 today at the ROSCon conference in Toronto, the latest version of its GPU-accelerated package collection built on the open-source Robot Operating System (ROS), adding AI agent support aimed at speeding up how developers build and deploy robots.

What's new

Isaac ROS 5.0 adds support for ROS Lyrical and Ubuntu 24.04, and NVIDIA worked with the Open Source Robotics Alliance to contribute a standard data-handling interface to ROS Lyrical that helps robotics software run efficiently across different computing hardware, including GPUs.

The release's headline addition is a set of "agent-ready" skills and documentation meant to let AI agents, not just human developers, carry out robotics development tasks:

• A new FoundationStereo fine-tuning skill lets an AI agent adapt a stereo perception model to a developer's own cameras and environment.
• FoundationPose, NVIDIA's foundation model for object pose estimation and tracking, now ships an agent-ready inference library that the company says tracks object position and orientation up to 5.5x faster.
• A pick-and-place workflow — chaining detection, depth estimation and pose output — is now available as a standalone, agent-ready skill.

NVIDIA Isaac ROS — a collection of GPU-accelerated packages built on ROS — is released today at the ROSCon conference in Toronto, Canada, and is described by NVIDIA as helping "humans and AI agents build robots together." The company says Isaac ROS brings its accelerated computing, physical AI models and production-ready libraries to what it counts as nearly 1.3 million ROS users. The release is free and open source, available now on GitHub.

Context

ROS is the open framework, maintained by Open Robotics, that much of the robotics industry builds on for common tools, libraries and standards. NVIDIA has spent several release cycles adding GPU-accelerated perception and navigation packages on top of that base through Isaac ROS. Isaac ROS 5.0 extends that work into the current wave of coding agents: the ecosystem move is already visible in AgenticROS, an open-source project sponsored by 3D-perception company RealSense that connects Isaac ROS with NVIDIA's Nemotron open models and NemoClaw blueprints so AI agents can interact directly with ROS-based robots.

Why it matters

The release is a concrete example of AI coding-agent tooling moving from software development generally into a more specialized, hardware-adjacent domain: robotics, where mistakes have physical consequences and perception pipelines are harder to get right than typical application code. By packaging fine-tuning and inference tasks as reusable "skills" with agent-ready documentation, NVIDIA is betting that agentic development — already common for web and backend code — can meaningfully cut the time it takes to adapt perception models to new sensors and environments, a task that has traditionally required specialist robotics engineers. Because Isaac ROS remains free and open source, the release also reinforces NVIDIA's strategy of driving hardware demand (Jetson edge AI platforms, GPU acceleration) by keeping the software layer on top of it open and widely adopted.