Skip to content
← All builds
Try NVIDIA Isaac GR00T, an open humanoid foundation model: media published by NVIDIA Isaac GR00T
Demo video on YouTube (opens in a new tab)Credit: NVIDIA Isaac GR00T
Try NVIDIA Isaac GR00T, an open humanoid foundation model: media published by NVIDIA Isaac GR00T (image 2)
Try NVIDIA Isaac GR00T, an open humanoid foundation model: media published by NVIDIA Isaac GR00T (image 3)

Try NVIDIA Isaac GR00T, an open humanoid foundation model

Humanoid, Hard

Isaac GR00T is NVIDIA's open vision-language-action foundation model for humanoids and arms, with weights on Hugging Face and fine-tuning scripts.

View build on GitHub ↗Share on X

Fine-tune GR00T on LeRobot-format data (the repo has an examples/SO100 walkthrough) and deploy. Datasets in the Physical AI collection help you start.

Hardware & cost: An NVIDIA GPU; any robot with demo data.

Difficulty: Hard.

Agent install prompt (paste into Claude Code, Cursor or Codex): "Clone https://github.com/NVIDIA/Isaac-GR00T, set up the env per the README, run the inference example on a sample dataset, then show me the fine-tune command for an SO-101 dataset."

Source: NVIDIA/Isaac-GR00T on GitHub — https://github.com/NVIDIA/Isaac-GR00T (facts checked against the project README, Oct 2026; prices vary by region).

Quick answers

What do you need to try NVIDIA Isaac GR00T, an open humanoid foundation model?
An NVIDIA GPU; any robot with demo data.
How hard is it to try NVIDIA Isaac GR00T, an open humanoid foundation model?
Hard.

How your agent drives it

Copied from the NVIDIA/Isaac-GR00T README

  1. Shell
    $ git clone --recurse-submodules https://github.com/NVIDIA/Isaac-GR00T
  2. Shell
    $ uv sync --python 3.12
  3. Shell
    $ uv run python scripts/deployment/standalone_inference_script.py \
    --model-path nvidia/GR00T-N1.7-3B \
    --dataset-path demo_data/droid_sample \
    --embodiment-tag OXE_DROID_RELATIVE_EEF_RELATIVE_JOINT \
    --traj-ids 1 2 \
    --inference-mode pytorch \
    --execution-horizon 8

Any coding agent that can run a shell can run these. Read the README before you run anything on a robot that moves.

Ideas to try with it

Our ideas, not the builder's claims.

  1. Run inference on the sample DROID data before collecting your own
  2. Fine-tune on SO-101 demos and compare it to a smaller model
  3. Convert a LeRobot dataset and train GR00T on it
1 more idea
  1. Test it on a Jetson for on-robot inference
Source
NVIDIA/Isaac-GR00T on GitHub
Last commit
2026-10-29
Builder
NVIDIA
Repo
NVIDIA/Isaac-GR00T, 8,309 stars
Built with
SO-100/SO-101 arm (LeRobot), Unitree G1

More like this: SO-101 / LeRobot arms

What to build next

Builds that share a board, robot or agent with this one.