# NVIDIA and Hugging Face Put an Open Robot Foundation Model Into LeRobot

Isaac GR00T 1.7, a teleoperation framework, and a coming world model land in the open-source robotics stack.

- Published: 2026-07-07T10:40:11.754Z
- Canonical: https://polylog.news/ai/2026-07-07/nvidia-and-hugging-face-put-an-open-robot-foundation-model-i
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [NVIDIA Blog](https://blogs.nvidia.com/blog/hugging-face-lerobot-models-frameworks-open-robotics/), [Hugging Face](https://huggingface.co/blog/nvidia/gr00t-n1-7)

NVIDIA and Hugging Face are moving robotics toward the open-model workflow that has defined language AI. The two [brought Isaac GR00T 1.7](https://blogs.nvidia.com/blog/hugging-face-lerobot-models-frameworks-open-robotics/), described as an open, commercially usable reasoning vision-language-action (VLA) model for humanoids, into LeRobot, Hugging Face's open robotics library, alongside the Isaac Teleop framework for capturing human demonstrations in standardized formats.

The design target is post-training and embodiment transfer: developers adapt GR00T to new robot bodies and tasks through LeRobot pipelines, then deploy on hardware including [NVIDIA's Jetson Thor and the open Reachy 2 humanoid](https://huggingface.co/blog/nvidia/gr00t-n1-7). NVIDIA also flagged Cosmos 3, a world foundation model for generating and augmenting robotics data, as coming to the ecosystem to reduce reliance on expensive real-world data collection.

The persistent constraint in embodied AI is data, not model architecture, and this release attacks that directly by standardizing demonstration collection and adding synthetic-data generation. The open commercial license is the strategic move: it makes GR00T a default substrate that keeps physical-AI development anchored to NVIDIA's compute and tooling even as the weights circulate freely.

## What this means

Robotics is following the language-model playbook, where an openly downloadable foundation model plus shared tooling compounds developer contributions faster than closed suites. For NVIDIA, giving away the model while owning the training and deployment silicon is a strategy to convert an open ecosystem into recurring chip demand.

## What to watch

- Whether independent labs report successful GR00T transfer to non-reference robot bodies, the real test of a general robot foundation model versus a demo tied to one platform.
- Adoption of Isaac Teleop's data formats by rival robot makers, which would signal an emerging open standard for embodied-AI training data.