NVIDIA released Isaac ROS 5.0 on September 22, a new version of its toolkit for developing robots based on ROS and GPUs.
The release adds ROS Lyrical and Ubuntu 24.04 support, broadens Jetson family support and brings tools and documentation into development designed to work with AI agents too.
Unlike many "Physical AI" announcements remaining demonstrations, Isaac ROS 5.0 is available now and free.
The launch also has an Israeli angle: Mentee Robotics uses Isaac ROS and Jetson Thor in developing MenteeBot V3.
What is Isaac ROS?
ROS, or Robot Operating System, is a widely used open ecosystem for robot software development.
Isaac ROS is NVIDIA's tool layer adding GPU acceleration and ready-made packages for tasks such as visual perception, localization, sensor processing and manipulation.
The goal is not to replace ROS, but to let developers use NVIDIA hardware and libraries to accelerate parts of the work.
Isaac ROS 5.0 continues this direction but adds a new layer: adaptation to workflows where AI agents participate in development.
What does "agent-ready" mean?
The term can sound as though a robot starts programming itself.
That is not what NVIDIA presents.
The company adds documentation, workflows and skills intended to let development agents better understand the environment and use existing tools.
According to NVIDIA, capabilities include setup tasks, manipulation, FoundationStereo fine-tuning and pick-and-place processes.
In other words, an AI agent can help a developer build, configure and modify a robotic pipeline, but still works within a tool and permissions system someone else defined.
FoundationPose is faster, according to NVIDIA
One improvement emphasized is FoundationPose.
NVIDIA presents a new inference library adapted for agent work and claims up to 5.5 times faster performance.
As always with such a figure, examine measurement conditions.
The research currently contains no complete external benchmark establishing that a 5.5-fold improvement applies to every hardware, model or robotics task.
This is NVIDIA's figure and should be treated as such.
Broader Jetson support
The new version is intended to run on hardware ranging from Jetson Orin Nano to Jetson Thor.
This matters because different robotics projects need different performance levels.
A small prototype or lightweight system can run on cheaper, more efficient hardware, while a humanoid robot or complex perception system may require much more powerful hardware.
However, "support" does not mean every workflow runs at the same speed on every device.
Developers still need to adapt models, processing rates and sensors to hardware limits.
The connection to Israel's Mentee Robotics
NVIDIA presents Israel's Mentee Robotics as one company using Isaac ROS and Jetson Thor.
Mentee develops MenteeBot V3, a humanoid robot.
This connection gives the launch a genuine Israeli angle, not merely general robotics-market context.
However, the announcement contains no data establishing that the robot is already deployed commercially at scale or allowing business results to be inferred from Isaac ROS use.
Open source does not mean no NVIDIA dependence
Isaac ROS is based on the open ROS ecosystem, and some packages are open source.
But using Isaac ROS still connects a project deeply to NVIDIA technologies.
CUDA, Jetson and dedicated library acceleration can provide a performance advantage but also increase vendor dependence.
An organization choosing this stack needs to consider not just performance but hardware availability, costs, moving the project to another platform and how much code depends on NVIDIA.
What an AI agent can help do
The potential advantage of combining agents and robotics is not that an agent "understands a robot" like an engineer.
It is the ability to work faster with documentation, code, configurations and tools.
An agent can help a developer find a suitable example, generate configuration files, adapt a pipeline, analyze an error or make repetitive changes.
But in robotics, mistakes do not stay on the screen.
Incorrect software can move an arm, damage an object or cause unexpected physical system behavior.
Approval and real-hardware testing therefore remain critical.
What remains unknown
The release is very recent, and most initial technical information comes from NVIDIA.
There is no broad independent testing base for version 5.0 yet.
For example, the exact FoundationPose improvement measurement conditions, how much working time the new skills save in a real project and which workflows are production-ready rather than examples or reference implementations remain unclear.
These are things the ROS community and developers will need to examine soon.
Why it matters
The connection between generative AI and robotics is advancing rapidly from two directions.
On one side, models gain more advanced perception and reasoning capabilities.
On the other, AI agents enter the process by which developers themselves build and maintain robots.
Isaac ROS 5.0 illustrates the second direction.
It does not make robots fully autonomous, but brings agentic coding closer to the physical world.
This is a change robotics developers, including Israeli companies, will need to follow closely.