InternUtopia Documentation

InternUtopia is a research platform for embodied AGI featuring three main pillars:

• GRScenes: A scene dataset containing 100k interactive, finely annotated scenes across 89 diverse categories.
• GRResidents: An LLM-driven NPC system that enables social interaction, task generation, and task assignment to simulate social scenarios.
• GRBench: A collection of embodied AI benchmarks for assessing capabilities in tasks like Object Loco-Navigation, Social Loco-Navigation, and Loco-Manipulation.

Before installing InternUtopia, ensure your system meets the following requirements:

• OS: Ubuntu 20.04 or 22.04
• Simulator: NVIDIA Omniverse Isaac Sim 4.5.0
  • Requires an NVIDIA GPU (RTX 2070 or higher)
  • Recommended NVIDIA GPU Driver version: 535.216.01+
  • Docker and NVIDIA Container Toolkit are optional but supported.
• Environment: Conda with Python 3.10.16 (3.10.* is acceptable).

After installation, you must download the required assets.

Important: You MUST first complete the User Agreement for GRScenes-100 Dataset Access before proceeding.

To download assets (either the full ~80GB set or a minimum ~500MB set), run:

python -m internutopia.download_assets

By default, assets are stored in ${PATH/TO/INTERNUTOPIA/ROOT}/internutopia/assets.

To ensure scenes load materials correctly, you must configure the MDL_SYSTEM_PATH environment variable to include the path to the Materials folder found within your downloaded GRScenes directory (e.g., GRScenes-100/home_scenes/Materials). It is recommended to add this to your ~/.bashrc.

# step 1. Find the Materials folder path in the downloaded GRScenes folder, such as `GRScenes-100/home_scenes/Materials`
# step 2. Configure the environment variable `MDL_SYSTEM_PATH` (saved to `~/.bashrc` is recommended).
echo 'export MDL_SYSTEM_PATH=$MDL_SYSTEM_PATH:</path/to/your_downloaded_materials_folder>' >> ~/.bashrc
source ~/.bashrc

Once Infinigen is installed and assets are applied, you can generate indoor scenes. Follow the standard Infinigen HelloRoom.md procedure, but ensure you are using the GRScenes-100 static assets you prepared.

Important: When executing generation commands, use Blender's bundled Python interpreter located at: infinigen/blender/4.2/python/bin/python3.11.

If you prefer not to use the built-in download script, you can manually download components from the following sources:

GRScenes-100

Download scene assets from:

• OpenDataLab
• ModelScope
• HuggingFace

Robots & Policy Weights

Download the robot directory from:

• OpenDataLab
• ModelScope
• HuggingFace

Note: After manual download, move the robot directory to the root of your configured asset path.

To use the procedural indoor scene generation workflow, you must first install the Infinigen framework. It is recommended to use a dedicated Conda environment with Python 3.11 and the interactive Blender installation script.

git clone https://github.com/princeton-vl/infinigen.git
cd infinigen
conda create --name infinigen python=3.11
conda activate infinigen
bash scripts/install/interactive_blender.sh

The scripts require OpenUSD and the IsaacSim Python SDK. If you are not using an existing GRUtopia conda environment, follow these steps to create a new environment and install the necessary packages from the NVIDIA PyPI index.

conda create -n <env_name> python=3.10
conda activate <env_name>
pip install usd-core==24.11
pip install isaacsim==4.2.0.2 isaacsim-extscache-physics==4.2.0.2 isaacsim-extscache-kit==4.2.0.2 isaacsim-extscache-kit-sdk==4.2.0.2 --extra-index-url https://pypi.nvidia.com

To optimize the GRScenes-100 data for use in Isaac Sim, run the data_post_processing.py script within the InternUtopia environment. This will produce a final_scene.usd file.

Output Location: The resulting final_scene.usd will be located in the {{path_to_infinigen}}/outputs/my_export/export_scene.blend directory.

python toolkits/indoor_scenes_generation/util/data_post_processing.py \
    --input_path {{path_to_infinigen}}/outputs/my_export/export_scene.blend/export_scene.usdc \
    --source_assets {{path_to_infinigen}}/infinigen/assets/static_assets/source

Convert your generated .blend files into the .usdc format using the Infinigen export tool.

python -m infinigen.tools.export \
     --input_folder {PATH_TO_FOLDER_OF_BLENDFILES} \
     --output_folder outputs/my_export \
     -f usdc \
     -r 1024