Skip to content

🛠️ Installation

🗒️ Requirements

Please make sure your system meets the following specs:

  • OS: Ubuntu 20.04+ / Windows 10+
  • RAM: 32GB+
  • GPU: NVIDIA RTX 2070+
  • VRAM: 8GB+
Why these specs?

OmniGibson is built upon NVIDIA's Omniverse and Isaac Sim platforms, so we inherit their dependencies. For more information, please see Isaac Sim's Requirements.

💻 Setup

There are three ways to setup OmniGibson, all built upon different ways of installing NVIDIA Isaac Sim:

  • 🐍 Install with pip (Linux / Windows, Recommended): You can install Omnigibson and automatically install Isaac Sim through pip for the fastest startup.
  • 🐳 Install with Docker (Linux only): You can quickly get OmniGibson immediately up and running from our pre-built docker image that includes Isaac Sim.
  • 🧪 Install with Omniverse Launcher (Linux / Windows): You can install Isaac Sim via the Omniverse launcher and hook OmniGibson up to it.

  1. Create a conda environment with Python version 3.10 and numpy and PyTorch:

    conda create -n omnigibson python=3.10 pytorch torchvision torchaudio pytorch-cuda=12.1 "numpy<2" -c pytorch -c nvidia
conda activate omnigibson
    What should I do if conda create fails?

    Sometimes, conda will fail to resolve dependencies. In that case, you can create a Python-only conda environment first, and then install numpy<2 and PyTorch via pip.

    If the default PyTorch version does not work for you due to a CUDA version compatibility issue, follow instructions on the PyTorch website to add the correct index option to the pip install line to get a different pytorch version.

    conda create -n omnigibson python=3.10
conda activate omnigibson
pip install "numpy<2" torch torchvision torchaudio

  2. Install OmniGibson:

    Install from PyPI (source not editable)

    pip install omnigibson

    Install from GitHub (source editable)

    git clone https://github.com/StanfordVL/OmniGibson.git
cd OmniGibson
pip install -e .

    Nightly build

    The main branch contains the stable version of OmniGibson. For our latest developed (yet not fully tested) features and bug fixes, please clone from the og-develop branch.

  3. Run the installation script to install Isaac Sim as well as OmniGibson dataset and assets:

    python -m omnigibson.install

    You can apply additional flag --no-install-datasets to skip dataset install.

    If this step fails, we recommend considering the source installation method.

More information

For more details on installing Isaac Sim via pip, please refer to the official Isaac Sim documentation.

Install OmniGibson with Docker is supported for 🐧 Linux only.

Need to install docker or NVIDIA docker? 
# Install docker
curl https://get.docker.com | sh && sudo systemctl --now enable docker

# Install nvidia-docker runtime
distribution=$(. /etc/os-release;echo $ID$VERSION_ID) \
    && curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey | \
    sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg \
    && curl -s -L https://nvidia.github.io/libnvidia-container/$distribution/libnvidia-container.list | \
    sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' | \
    sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list
sudo apt-get update
sudo apt-get install -y nvidia-docker2 # install
sudo systemctl restart docker # restart docker engine

  1. Install our docker launching scripts: 

    curl -LJO https://raw.githubusercontent.com/StanfordVL/OmniGibson/main/docker/run_docker.sh
chmod a+x run_docker.sh

    What is being installed?

    Our docker image automatically ships with a pre-configured conda virtual environment named omnigibson with Isaac Sim and OmniGibson pre-installed. Upon running the first time, our scene and object assets will automatically be downloaded as well.

  2. Then, simply launch the shell script:

    sudo ./run_docker.sh -h <ABS_DATA_PATH> # (1)!
    1. <ABS_DATA_PATH> specifies the absolute path data will be stored on your machine (if no <ABS_DATA_PATH> is specified, it defaults to ./omnigibson_data). This needs to be called each time the docker container is run!
    sudo ./run_docker.sh <ABS_DATA_PATH> # (1)!
    1. <ABS_DATA_PATH> specifies the absolute path data will be stored on your machine (if no <ABS_DATA_PATH> is specified, it defaults to ./omnigibson_data). This needs to be called each time the docker container is run!
    Are you using NFS or AFS?

    Docker containers are unable to access NFS or AFS drives, so if run_docker.sh are located on an NFS / AFS partition, please set <DATA_PATH> to an alternative data directory located on a non-NFS / AFS partition.

Install OmniGibson with Omniverse Launcher is supported for both 🐧 Linux (bash) and 📁 Windows (powershell/cmd).

  1. Install Conda and NVIDIA's Omniverse Isaac Sim

    Please make sure you have the currently supported version of Isaac Sim (4.1.0) installed.

    For Ubuntu 22.04, you need to install FUSE to run the Omniverse Launcher AppImage.

  2. Create a conda environment with Python version 3.10:

    conda create -n omnigibson python=3.10 pytorch torchvision torchaudio pytorch-cuda=12.1 "numpy<2" -c pytorch -c nvidia
conda activate omnigibson
    What should I do if conda create fails?

    Sometimes, conda will fail to resolve dependencies. In that case, you can create a Python-only conda environment first, and then install numpy<2 and PyTorch via pip.

    If the default PyTorch version does not work for you due to a CUDA version compatibility issue, follow instructions on the PyTorch website to add the correct index option to the pip install line to get a different pytorch version.

    conda create -n omnigibson python=3.10
conda activate omnigibson
pip install "numpy<2" torch torchvision torchaudio

  3. Install OmniGibson:

    Install from PyPI (source not editable)

    pip install omnigibson

    Install from GitHub (source editable)

    git clone https://github.com/StanfordVL/OmniGibson.git
cd OmniGibson
pip install -e .

    Nightly build

    The main branch contains the stable version of OmniGibson. For our latest developed (yet not fully tested) features and bug fixes, please clone from the og-develop branch.

  4. Run the installation script to hook the environment up to Isaac Sim as well as OmniGibson dataset and assets:

    python -m omnigibson.install --launcher-install

    You can specify your Isaac Sim install location using the argument --isaac-sim-path if it differs from the default. You can also apply additional flag --no-install-datasets to skip dataset install.

    What does this do?

    When you install OmniGibson this way, it will modify your conda environment setup to hook it up to the launcher-installed Isaac Sim.

  5. Deactivate and reactivate the conda environment:

    Because the environment was modified by the installer to hook it up to the launcher-installed Isaac Sim, you need to reactivate it.

    conda deactivate
conda activate omnigibson

🌎 Explore OmniGibson!

Expect slowdown during first execution

Omniverse requires some one-time startup setup when OmniGibson is imported for the first time.

The process could take up to 5 minutes. This is expected behavior, and should only occur once!

OmniGibson is now successfully installed! You can try teleoperating one of our robots:

python -m omnigibson.examples.robots.robot_control_example --quickstart # (1)!
  1. This demo lets you choose a scene, robot, and set of controllers, and then teleoperate the robot using your keyboard. The --quickstart flag will automatically select the scene and robot for you - remove that if you want to change the scene or robot.

You can also try exploring some of our new scenes interactively:

python -m omnigibson.examples.scenes.scene_selector # (1)!
  1. This demo lets you choose a scene and interactively move around using your keyboard and mouse. Hold down Shift and then Left-click + Drag an object to apply forces!

Next: Get quickly familiarized with OmniGibson from our Quickstart Guide!

🧯 Troubleshooting

I cannot open Omniverse Launcher AppImage on Linux

You probably need to install FUSE to run the Omniverse Launcher AppImage.

OmniGibson is stuck at HydraEngine rtx failed creating scene renderer.

OmniGibson is likely using an unsupported GPU (default is id 0). Run nvidia-smi to see the active list of GPUs, and select an NVIDIA-supported GPU and set its corresponding ID when running OmniGibson with export OMNIGIBSON_GPU_ID=<ID NUMBER>.

I'm getting the error AttributeError: module 'x509' has no attribute 'Store'

This happens because launcher-based versions of Isaac Sim ship with a faulty copy of the cryptography module. You can simply look at the stack trace for the path of your copy of the cryptography module, e.g. if it shows an error that looks like this:

File "c:/users/cem/appdata/local/ov/pkg/isaac-sim-4.1.0/exts/omni.pip.cloud/pip_prebundle/cryptography/hazmat/backends/openssl/__init__.py", line 7, in <module>
    from cryptography.hazmat.backends.openssl.backend import backend
File "c:/users/cem/appdata/local/ov/pkg/isaac-sim-4.1.0/exts/omni.pip.cloud/pip_prebundle/cryptography/hazmat/backends/openssl/backend.py", line 12, in <module>
    from cryptography import utils, x509
File "c:/users/cem/appdata/local/ov/pkg/isaac-sim-4.1.0/exts/omni.pip.cloud/pip_prebundle/cryptography/x509/__init__.py", line 7, in <module>
    from cryptography.x509 import certificate_transparency, verification
File "c:/users/cem/appdata/local/ov/pkg/isaac-sim-4.1.0/exts/omni.pip.cloud/pip_prebundle/cryptography/x509/verification.py", line 20, in <module>
    Store = rust_x509.Store
AttributeError: module 'x509' has no attribute 'Store'

You can simply remove the c:/users/cem/appdata/local/ov/pkg/isaac-sim-4.1.0/exts/omni.pip.cloud/pip_prebundle/cryptography directory and this issue should be resolved.