On July 8, 2026, NVIDIA and Hugging Face announced the integration of Isaac GR00T 1.7 and the Isaac Teleop framework into LeRobot. The move connects NVIDIA's 3 million robotics developers with Hugging Face's 16 million AI builders. Cosmos 3 — a frontier world model for physical AI — is the announced next step.
Key takeaways
- Isaac GR00T 1.7 — the first open and commercially viable VLA model for humanoid robots — joins LeRobot
- Isaac Teleop enables standardized collection and community sharing of human demonstration data
- LeRobot already hosts over 15 million dataset downloads and more than 350,000 trajectories
- Cosmos 3 (frontier world model for physical AI) announced for the platform — no release date given
- Jetson Thor integration supports VLA model deployment on the Reachy 2 humanoid from Pollen Robotics
Two communities, one toolchain
For years, building robots required separate ecosystems. AI researchers worked in PyTorch and Hugging Face Hub. Robotics developers relied on ROS, Isaac Sim, and dedicated simulators. They rarely shared tools, data formats, or models.
The integration of GR00T 1.7 and Isaac Teleop into LeRobot reshapes that landscape. LeRobot is an open library for training and sharing robot policies, datasets, and workflows. NVIDIA developers can now work directly in the same environment as the Hugging Face community.
"Open source is how a field turns advanced research into something people can study, adapt and build on. With NVIDIA Isaac GR00T 1.7 and Isaac TeleOp in LeRobot today, robotics developers can use shared models, data, and workflows to train and evaluate robots in the open."
— Thomas Wolf, co-founder and chief science officer, Hugging Face
What GR00T 1.7 and Isaac Teleop add
Isaac GR00T 1.7 is a vision-language-action model for humanoid robots. NVIDIA describes it as the first open and commercially viable robot foundation model. Earlier models of this class were either proprietary or purely experimental.
Within LeRobot, GR00T 1.7 can be fine-tuned for new robot embodiments and tasks with benchmarked performance. Before this integration, a developer had to build a custom pipeline — from data conversion to optimizer selection and validation. All those steps now share a common interface.
Isaac Teleop handles the data side. It is an open-source framework for collecting human demonstrations via external devices. Output follows standardized, interoperable formats and can be shared directly within the LeRobot community.
Both tools target the same bottleneck: cost and fragmentation of physical AI resources. Until now, training a humanoid policy required expensive datasets and separate tools from multiple vendors.
The LeRobot ecosystem — what's already inside
The GR00T 1.7 and Teleop integrations build on capabilities LeRobot already offers. The library hosts a large public physical AI dataset, downloaded more than 15 million times, containing over 350,000 real and simulated trajectories and 57 million grasps.
On the simulation side, LeRobot already includes environments based on Isaac Sim and Isaac Lab. These allow developers to configure environments, generate data, and test policies before moving to physical hardware. A new addition is Isaac Lab-Arena in the LeRobot Environment Hub — enabling rapid prototyping of complex simulation environments and training of generalist models such as GR00T, Pi, and SmolVLA.
At the deployment level, Jetson Thor integration with Reachy 2 — the open-source humanoid from Pollen Robotics — provides a ready path from VLA model to physical execution on an open humanoid platform.
Cosmos 3 — announced, not yet available
The companies announced the upcoming integration of Cosmos 3 into LeRobot. Cosmos 3 is a frontier world foundation model for physical AI, designed to help developers generate and augment robotics data, simulate scenarios, and support policy development when real-world data is scarce or costly.
At this stage, the integration is an announcement without a timeline, version details, or access terms.
Why this matters
Physical AI — the ability of a robot to act in the real world based on a trained model — remains expensive. Data is fragmented, tools are inconsistent, and foundation models are hard to access. Most robotics teams start from scratch on every project.
The NVIDIA-LeRobot integration lowers that barrier measurably. A developer can take GR00T 1.7 as a starting point, collect demonstration data through Teleop in a standard format, fine-tune the model in LeRobot, and deploy it on a physical platform — all within one workflow.
The scale matters just as much: 19 million developers in a single shared ecosystem. Open-source models mature quickly with large user bases. The transformer breakthrough in NLP owed its speed not to a single architecture, but to openness to iteration across thousands of teams. Robotics is following the same path, with a lag of a few years. If Cosmos 3 reaches LeRobot as announced, it will add a layer of synthetic data that could cut the time to a working prototype by months.
What's next
- NVIDIA has announced the Cosmos 3 integration with LeRobot — no timeline or technical details provided yet
- GR00T 1.7 is available in the Hugging Face Hub and deployable through LeRobot workflows as of July 8, 2026
- Developers can test the Isaac Lab-Arena integration through LeRobot EnvHub today
Sources
- The Robot Report — NVIDIA and Hugging Face bring new models and frameworks to LeRobot
- NVIDIA Developer Blog — Develop Humanoid Robot Policies End-to-End with NVIDIA Isaac GR00T
- GitHub — NVIDIA Isaac Teleop
