NVIDIA & Hugging Face add robotics tools to LeRobot

NVIDIA and Hugging Face have added NVIDIA Isaac GR00T 1.7 and the NVIDIA Isaac Teleop framework to the LeRobot open-source robotics library, extending their work on shared tools for robot development.

The additions bring two NVIDIA technologies into LeRobot, Hugging Face's library for training, running and sharing robot datasets, models, policies and workflows. Cosmos 3, NVIDIA's world foundation model for physical AI, is also due to be added later.

The collaboration focuses on open robotics development, as companies try to lower the cost and complexity of building and testing AI systems for physical machines. By linking model training, data collection and evaluation tools in a common software environment, the two groups aim to make it easier for developers to work across different robots and tasks.

Isaac Teleop is designed for robot data collection through human demonstrations captured from external devices in standardized formats. In LeRobot, developers can use those demonstrations to build datasets and share them with other users.

NVIDIA describes Isaac GR00T 1.7 as an open reasoning vision-language-action model for humanoid robots. The model can be post-trained and deployed through LeRobot workflows, allowing developers to adapt it for different robot designs and tasks.

Cosmos 3, which has not yet been added to the library, is intended to support robotics teams when real-world data is scarce or expensive to collect. It is designed to generate and augment data, simulate scenarios and support policy development.

Open workflows

The wider partnership also brings together two large developer communities. NVIDIA says it has 3 million robotics developers, while Hugging Face says 16 million AI builders use its platform and tools.

The companies argue that a shared framework for data, models and testing could reduce fragmentation in robotics development. Teams often work with separate simulation software, training pipelines and validation methods, making it harder to compare results or reuse work across projects.

LeRobot has become one of the more prominent open-source libraries for robotics developers who want to share models, datasets and workflows in a common format. Bringing NVIDIA's software into that environment gives developers access to data collection tools and a robot foundation model without requiring a separate stack for those parts of the process.

Thomas Wolf, Co-Founder and Chief Science Officer at Hugging Face, said the integration reflects the role of open source in robotics development.

"Open source is how a field turns advanced research into something people can study, adapt and build on," Wolf said.

He added: "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. And with NVIDIA Cosmos 3 planned next, the community will have a path to bring frontier world models into that same collaborative loop."

Broader stack

The latest additions build on a broader set of NVIDIA resources already connected to LeRobot. These include a physical AI dataset that, according to NVIDIA, has been downloaded more than 15 million times and contains more than 350,000 real and simulated trajectories, as well as 57 million grasps.

NVIDIA has also linked LeRobot with Isaac Sim and Isaac Lab, which are used to set up environments, generate robot data, test policies and validate behavior before systems are moved to physical robots. Another component, Isaac Lab-Arena, sits in the LeRobot Environment Hub and is intended to help developers prototype simulation environments and use them in training and evaluation.

There is also an integration between NVIDIA Jetson Thor and LeRobot's Reachy 2 for deploying vision-language-action models on open-source humanoid robots. Together, the integrations show how NVIDIA is using open-source software communities to widen access to its robotics tools while placing its models and frameworks inside a workflow already used by developers sharing code, data and experiments.