NVIDIA and Hugging Face add Isaac robotics tools to LeRobot

NVIDIA says it is working with Hugging Face to bring more of its physical AI stack into LeRobot, the open source robotics library used to train, run and share robot models, datasets and workflows. The July 6 announcement adds NVIDIA Isaac GR00T 1.7 and Isaac Teleop to LeRobot, with NVIDIA Cosmos 3 planned for a later integration.

The move matters because robot development is still split across expensive datasets, simulation systems, model tooling and deployment pipelines. By putting more of those pieces into a shared open workflow, the companies are trying to make it easier for researchers and developers to collect demonstrations, fine-tune robot foundation models, test behavior in simulation and deploy policies on different robot bodies.

What developers get

According to NVIDIA, Isaac Teleop gives developers an open framework for collecting robot training data from human demonstrations, using standardized formats that can be expanded and shared through LeRobot. Isaac GR00T 1.7 is described as an open, commercially viable vision-language-action model for humanoid robots that can be post-trained and deployed through LeRobot workflows.

• LeRobot becomes a common path for data collection, training, evaluation and deployment.
• GR00T 1.7 can be adapted to new robot embodiments and tasks through post-training.
• Cosmos 3 is planned to help generate and augment robotics data, simulate scenarios and support policy development when real-world data is limited.

NVIDIA also pointed to existing resources already tied into the ecosystem, including open physical AI datasets with more than 350,000 real and simulated trajectories and 57 million grasps, as well as Isaac Sim and Isaac Lab-based simulation tools. The company says its robotics developer base numbers about 3 million, while Hugging Face reaches about 16 million AI builders.

Hugging Face's own GR00T N1.7 material, published earlier this year, describes the model as a 3-billion-parameter open reasoning vision-language-action system for humanoid robots, available for commercial and non-commercial use. That earlier release supplied the model and code; this week's LeRobot work is about making the model and supporting data-collection tools easier to use inside a broader open robotics workflow.

For robotics teams, the practical test will be whether these integrations reduce the cost and friction of moving from demonstration data to reliable real-world behavior. For the open source robotics community, the announcement is another sign that foundation-model style development is moving from chat and coding into embodied systems.

Sources

• blogs.nvidia.com
• huggingface.co

Cover photo by Diego Martinez on Pexels, used under the Pexels License.