NVIDIA and Hugging Face open parts of a robot foundation-model stack, publish Isaac GR00T 1.7
- NVIDIA and Hugging Face said on July 7 they added Isaac GR00T N1.7 and Isaac Teleop to LeRobot, opening more of a shared robotics stack. - The companies said LeRobot has surpassed 15 million dataset downloads, while NVIDIA cited more than 350,000 robot trajectories across released workflows and datasets. - NVIDIA said Cosmos 3 integration is planned next, with GR00T N1.7, Isaac Teleop and related assets already available on Hugging Face.
NVIDIA and Hugging Face said on July 7 they had added NVIDIA Isaac GR00T N1.7 and Isaac Teleop to LeRobot, extending an open-source robotics toolkit that both companies are pitching as a shared development path for robot AI. The release gives developers access to a vision-language-action model, teleoperation workflows, datasets and training utilities through Hugging Face’s LeRobot framework. NVIDIA said a future LeRobot integration will also bring in Cosmos 3, its world-model family. ### What exactly did NVIDIA and Hugging Face release? Hugging Face said in a blog post published July 7 that Isaac Teleop and GR00T N1.7 are now available in LeRobot, alongside examples for data collection, training and evaluation. LeRobot is Hugging Face’s open-source robotics library for training, running and sharing robot datasets, models and policies. NVIDIA said GR00T N1.7 is an open, commercially licensed vision-language-action model for humanoid robots. (blogs.nvidia.com) The company’s GitHub repository describes N1.7 as the latest version of GR00T N1, with a new vision-language backbone and updated benchmarks, while Hugging Face documentation says the model is exposed in LeRobot through the `groot` policy type. ### Where does Isaac Teleop fit in this stack? (huggingface.co) NVIDIA said Isaac Teleop is the data-collection layer in the workflow, letting developers record human demonstrations and convert them into formats used by GR00T and LeRobot. The company’s technical documentation shows export paths into LeRobot datasets hosted on Hugging Face Hub, and the LeRobot repository added examples for Isaac Teleop-to-SO-101 teleoperation and dataset recording in recent updates. (huggingface.co) The July 7 NVIDIA technical post described the broader GR00T platform as an end-to-end workflow spanning simulation setup, teleoperation-based data collection, post-training and deployment. That matters because the announcement was not limited to a model checkpoint; it tied together data capture, dataset formatting, fine-tuning and inference inside one toolchain. (developer.nvidia.com) ### Why are the dataset and trajectory numbers getting so much attention? NVIDIA said the LeRobot community has passed 15 million dataset downloads, a figure it used to show how much robotics data is already circulating through the Hugging Face ecosystem. The company also said the integrated workflows are backed by more than 350,000 trajectories across toolchains and datasets released through the collaboration. (developer.nvidia.com) The LeRobot GitHub repository shows the project has more than 25,000 stars and frequent updates tied to GR00T N1.7 support, including new checkpoints and examples added this week. Those public repository changes provide a visible record that the integration is being maintained in the open rather than announced as a one-off drop. ### Is this fully open source, or only parts of the stack? NVIDIA said GR00T N1.7 is open and commercially licensed, but the companies described the release as opening access to models, workflows and datasets rather than open-sourcing every layer of NVIDIA’s robotics software. (blogs.nvidia.com) NVIDIA’s developer page still presents Isaac GR00T as a broader reference platform for building, training, testing and deploying robot policies. (github.com) Hugging Face’s documentation shows the practical effect for developers: GR00T can now be compared, fine-tuned and run inside the same LeRobot framework used for other robotics policies. That gives outside researchers and startups a common interface without requiring them to assemble all of NVIDIA’s infrastructure from scratch. ### What comes next in the rollout? NVIDIA said Cosmos 3 integration into LeRobot is planned next, adding what the company described as frontier world models to the same open robotics workflow. (huggingface.co) GR00T N1.7 model checkpoints, Isaac Teleop materials and associated datasets are already posted through NVIDIA’s GitHub and Hugging Face pages, where the companies are publishing subsequent updates. (blogs.nvidia.com) (huggingface.co)