NVIDIA and Hugging Face deepen robotics partnership, bringing GR00T 1.7 and Isaac Teleop to LeRobot

NVIDIA and Hugging Face are expanding their open robotics partnership, integrating NVIDIA's Isaac GR00T 1.7 foundation model and the new Isaac Teleop data-collection framework directly into Hugging Face's LeRobot library, with the Cosmos 3 physical-AI model planned to follow.

What's new

The centerpiece of the update is NVIDIA Isaac GR00T 1.7, which NVIDIA describes as its first open, commercially viable reasoning vision-language-action (VLA) model for humanoid robots, now wired into standard LeRobot workflows for post-training and deployment. Alongside it, NVIDIA is shipping Isaac Teleop, an open-source framework for capturing high-quality human demonstrations from external devices in standardized, interoperable formats, so developers can collect, expand, and share robot datasets without leaving LeRobot.

The companies say more is coming: NVIDIA Cosmos 3, described as a frontier model for physical AI, is planned for LeRobot "soon." The integrations build on infrastructure NVIDIA has already put into the open ecosystem, including what it calls the largest open-source physical-AI dataset -- more than 350,000 real and simulated robot trajectories and 57 million grasps -- plus Isaac Sim and Isaac Lab simulation frameworks for building training environments, and Isaac Lab-Arena, now available inside the LeRobot Environment Hub for prototyping simulation environments.

The timing lines up with Hugging Face's own LeRobot v0.6.0 release, which upgraded its NVIDIA GR00T integration to GR00T N1.7 -- swapping the model's previous vision-language backbone for Cosmos-Reason2-2B feeding into a flow-matching action head, part of a broader push toward "world model" robot policies that simulate outcomes before acting.

Context

LeRobot is Hugging Face's open-source library for training, running, and sharing robot datasets, models, and policies -- the robotics equivalent of what Transformers did for language models. NVIDIA has been building out an equivalent open stack on the hardware and simulation side for several years through Isaac Sim, Isaac Lab, and its GR00T model family. Tying the two together gives developers a single workflow that spans dataset collection, simulation, training, and deployment, rather than stitching together NVIDIA's tooling and Hugging Face's library separately.

Why it matters

NVIDIA frames the reach of the partnership in blunt numbers: connecting its roughly 3 million robotics developers with Hugging Face's 16 million AI builders. That's a bet that humanoid and embodied-AI development follows the same open-source dynamics that reshaped language models, where a shared, freely available foundation model and toolchain accelerate the whole field faster than any single company's closed stack. For a robotics industry still assembling its equivalent of the LLM open-weights moment, GR00T 1.7 landing as a standard LeRobot dependency, backed by NVIDIA's simulation and data infrastructure, pushes that moment closer. It also reinforces NVIDIA's strategy of winning the robotics era not just by selling chips, but by making its models and tools the default layer underneath everyone else's open-source robotics work.