Open, commercially licensed humanoid VLA (~3B) with Cosmos-Reason2 backbone and EgoScale human-video pretraining. Current GR00T line on Isaac / Hugging Face.
Best current open GR00T checkpoint for humanoid / NVIDIA-stack VLA work — commercially licensed and actively maintained.7-release — https://github.com/NVIDIA/Isaac-GR00T/releases/tag/n1.7-release.7-release — https://github.com/NVIDIA/Isaac-GR00T/releases/tag/n1.7-release.7-release — https://github.com/NVIDIA/Isaac-GR00T/releases/tag/n1.7-release. Latest GitHub release: n1.7-release — https://github.com/NVIDIA/Isaac-GR00T/releases/tag/n1.7-release.
Isaac GR00T N1.7 is NVIDIA’s open humanoid VLA line with commercial licensing, a Cosmos-Reason2 vision-language backbone, and large-scale human egocentric pretraining (EgoScale). Weights and reference code ship via Hugging Face and the Isaac-GR00T GitHub repo.
Teams building humanoid or cross-embodiment policies on the NVIDIA / Isaac stack who need an open, commercially usable VLA starting point.
Physical Intelligence generalist VLA (~3B) with flow-matching actions; π0 weights are partially open — strong dexterous-manipulation reference in the π0 family.
Google DeepMind closed VLA for whole-body control — from bipedal balance to fingertip dexterity — plus sibling ER / On-Device models in the Gemini Robotics 2 family.
Content reviewed 2026-08-02
