NVIDIA and Hugging Face expand LeRobot with Isaac GR00T 1.7 and Isaac Teleop

NVIDIA and Hugging Face deepened their open-robotics partnership by bringing a slate of new models and tools to the open-source LeRobot platform, aiming to democratize the physical-AI stack the way Hugging Face democratized language models. The headline additions are Isaac GR00T 1.7, a vision-language-action (VLA) foundation model purpose-built for humanoid robots, and the Isaac Teleop framework for teleoperated data collection and control.

The strategic logic is data. Robotics has lacked the massive open datasets that powered LLM progress, and VLA foundation models plus accessible teleoperation tooling are how the field bootstraps that data flywheel — humans demonstrate tasks via teleop, the demonstrations train the VLA model, and the model generalizes to new tasks. By putting GR00T 1.7 and Teleop into LeRobot's open ecosystem, NVIDIA seeds broad adoption of its robotics stack while Hugging Face becomes the hub for open robot models.

NVIDIA also signaled that support for Cosmos 3, its world foundation model, is planned — world models that let robots simulate and predict physical dynamics before acting. Combined with this week's Jetson Thor edge modules, NVIDIA now has an end-to-end robotics story: edge compute, VLA foundation models, teleop data tooling, and world models.

The community reaction was positive, with developers welcoming access to advanced robotics AI like Isaac GR00T for open-source projects. The caveat is that open robotics is still early — foundation models for embodiment lag their language counterparts by years, and Mistral's Robostral Navigate this week showed smaller, cheaper single-camera navigation models are a competing path. NVIDIA is betting the whole ecosystem consolidates on its tools.