Source: NVIDIA Blog — 2026-07-06
Summary
NVIDIA and Hugging Face announced new models and frameworks for LeRobot, the open-source robotics stack, including Isaac GR00T 1.7 — an open vision-language-action (VLA) model — and Isaac Teleop for collecting robot demonstration data. The companies also flagged planned integration of NVIDIA's Cosmos 3 world model into the LeRobot pipeline, aimed at lowering the barrier for the open robotics community to build and fine-tune physical-AI foundation models.
Key Takeaways
- Isaac GR00T 1.7 is released as an open VLA model usable directly within LeRobot's existing tooling, rather than a closed API.
- Isaac Teleop targets the data-collection bottleneck in robot learning — generating demonstration trajectories for downstream imitation learning.
- Cosmos 3 world-model integration is planned as a next step, suggesting a path toward simulation-augmented training within the same open stack.
- Signals continued NVIDIA investment in open-sourcing physical-AI foundation model components rather than keeping them proprietary.