The latest update to Nvidia's open-source robotics platform brings agentic workflows, faster object tracking, and support for nearly 1.3 million ROS users
Nvidia just gave its robotics development platform a major upgrade, and the centerpiece is something that would have sounded like science fiction a few years ago: AI agents that help build robots.
Isaac ROS 5.0, unveiled at the ROSCon conference in Toronto on September 22, introduces what Nvidia calls “agentic” development workflows. These are essentially AI-powered coding assistants purpose-built for robotics tasks, packaged alongside reusable procedures and documentation designed for both human developers and their AI counterparts.
What’s actually new
The headline feature is Nvidia’s new “Isaac agent skills,” which are modular building blocks for common robotics challenges. Think perception model tuning, object pose estimation, and pick-and-place workflows, the bread and butter of industrial robotics that traditionally require significant engineering time to get right.
On the performance side, the numbers are notable. A new agent-ready inference library for the FoundationPose model delivers object tracking speeds up to 5.5 times faster than previous versions.
There’s also a FoundationStereo fine-tuning skill, which lets developers optimize stereo perception models for their specific hardware setups and environments.
Isaac ROS 5.0 moves to the ROS 2 Lyrical distribution and runs on Ubuntu 24.04, keeping the platform current with the broader open-source robotics ecosystem. The entire release is available for free through GitHub and Nvidia’s Isaac ROS Buildfarm APT repository.
The agentic approach, explained
Nvidia’s approach packages robotics-specific knowledge into reusable skills that AI agents can invoke. Rather than asking a general-purpose AI to figure out how a robot arm should estimate the pose of an object, Isaac ROS 5.0 provides pre-built skills with comprehensive documentation that both human engineers and AI coding assistants can work with.
Scale and ecosystem play
Nvidia claims nearly 1.3 million ROS users currently leverage Isaac ROS in some capacity.
The update also integrates with Nvidia’s broader Isaac platform, including its Jetson edge hardware. Jetson processors are the workhorses powering autonomous machines from delivery robots to agricultural drones, so tighter integration between the development tools and the deployment hardware streamlines the path from prototype to production.
Beyond its own ecosystem, Nvidia is contributing a vendor-neutral memory transport interface to the ROS community through the Open Source Robotics Alliance.
