Skip to content
RobothecaBeta
Log in

Isaac GR00T N1.7

  • AI model
  • Humanoids and robot learning

An open vision-language-action model that turns camera images and language instructions into robot actions for manipulation and humanoid skills, post-trainable for new robots.

  • Custom weights terms

Made by NVIDIA, United States (checked on github.com/NVIDIA (opens in a new tab)).

Source on GitHub: github.com/NVIDIA/Isaac-GR00T (opens in a new tab). Facts read at commit dac54b3 (opens in a new tab).

Licenses

Code
Apache-2.0 (opens in a new tab)

Read by Robotheca from LICENSE (opens in a new tab) (Apache-2.0) at commit dac54b3 (opens in a new tab), on 16 Oct 2026.

Weights
NVIDIA LicenseTerms of their own, not a standard license

Read on huggingface.co/nvidia/GR00T-N1.7-3B/blob/main/LICENSE (opens in a new tab).

Commercial use
Unknown

Three statements disagree for the N1.7 weights: the README says Apache 2.0, the model card the NVIDIA Open Model License, and the model repository's LICENSE file is non-commercial. Unknown until NVIDIA clarifies.

Labels

  • Custom weights termsThe weights come under their maker's own terms (NVIDIA License), not a standard license: read their conditions before any use.

What it does

  • Cross-embodiment model taking language and image inputs to output continuous robot actions via a diffusion transformer head.
  • Built on a Cosmos-Reason2-2B (Qwen3-VL) vision-language backbone, replacing the Eagle backbone of N1.6.
  • Pretrained on robot demonstrations plus 20K hours of human video, using relative end-effector actions shared across embodiments.
  • Can be fine-tuned on custom robot data with launch_finetune.py and a modality config, with single- and multi-GPU support.
  • Offers zero-shot inference on pretrain embodiments and finetuned checkpoints for benchmarks such as LIBERO, DROID and SimplerEnv.
  • Supports deployment through a ZMQ policy server, open-loop and closed-loop evaluation, and ONNX/TensorRT export.

Written by Robotheca from the project's README (opens in a new tab) at commit dac54b3.

Getting started

Difficulty
Advanced
Hardware it needs, as the project states it
Inference needs 1 GPU with 16 GB+ VRAM (e.g., RTX 4090, L40, H100, Jetson AGX Thor/Orin, DGX Spark). Fine-tuning recommends 1 or more GPUs with 40 GB+ VRAM, with H100 or L40 nodes recommended. Platform setups cover dGPU on CUDA 12.8, Jetson Thor and Orin on JetPack 7.2, and DGX Spark on CUDA 13.0, a

Maintenance

Status
Active: a commit in the last 90 days
Newest commit
15 Oct 2026 on main
Latest release
n1.7 release (18 Apr 2026)

Checked on GitHub on 16 Oct 2026.

Links

Robots on Robotheca that use it

No robot on Robotheca is linked to it yet. Links come from a robot's robot.yaml, from its dependency files read as text, or from Robotheca's editors.

Robotheca lists AI models as a reference: nothing is hosted or run here, and every fact links to where it was read.