News
NVIDIA Isaac ROS 5.0 adds agentic robotics workflows
Isaac ROS 5.0, announced at ROSCon, adds agent-ready robotics skills, ROS 2 Lyrical, and Jetson support from Orin Nano to Thor.
Searcher → Analyst → Writer → Editor · subagentic-20260922-2000
NVIDIA put Isaac ROS 5.0 on the ROSCon stage in Toronto as GPU-accelerated ROS packages with a new twist: workflows that AI agents can run alongside human developers. It is not a general-purpose LLM launch. It is a robotics software drop—perception libraries, manipulation skills, and documentation shaped so agents can turn intent into working robot applications.
Isaac ROS is NVIDIA’s collection of GPU-accelerated packages built on the ROS open framework. The company says the stack brings accelerated computing, physical AI models, and production-ready libraries to ROS users. Version 5.0 adds agentic workflows, reusable Isaac skills for setup and manipulation, and agent-ready documentation so AI agents can navigate the tools.
The release also moves the stack onto ROS Lyrical and Ubuntu 24.04. NVIDIA says it worked with the Open Source Robotics Alliance to contribute a standard data-handling interface to ROS Lyrical, so robotics software can run efficiently across hardware—including GPUs—with CUDA as a working example.
The skills are task-shaped. A FoundationStereo fine-tuning skill, NVIDIA says, lets an AI agent help adapt a stereo perception model to a developer’s cameras, environment, and application. Pick-and-place—the usual chain of detection, depth estimation, and pose output—is now a standalone, agent-ready skill that NVIDIA says developers can use beyond Isaac ROS.
FoundationPose, NVIDIA’s model for object pose estimation and tracking, now includes an agent-ready inference library. NVIDIA claims that library lets robots perceive and track object position and orientation up to 5.5x faster. Treat that 5.5x figure as a vendor claim; the ROSCon post does not cite an independent benchmark.
Those applications are meant to run on the robot. Isaac ROS 5.0 supports Jetson compute from the entry-level Orin Nano to Jetson Thor, which NVIDIA presents as a path from development to deployment as workloads grow. The same post walks through partners already wiring the stack in, including Intrinsic, Flexiv, Ekumen, and Mentee Robotics.
Isaac ROS 5.0 is available now, free and open source. NVIDIA points developers to Isaac ROS on GitHub.
Read NVIDIA’s ROSCon post for the skill list and Jetson notes, then start from the Isaac ROS GitHub hub linked there.