Installation#
Docker is the recommended way to install AutoData. The dev container setup includes Isaac Sim, Isaac Lab, Isaac Lab-Arena, and AutoData, providing a reproducible environment without modifying the host Python installation. The repository is bind-mounted into the container, so edits on the host are live inside it.
An optional conda installation is also available for users who want direct control over their Python environment and installed packages. See Optional Conda Installation below.
Before installing, review the Support Matrix for the complete supported software stack, hardware requirements, optional cuRobo and XR dependencies, and resource guidance.
Common Prerequisites#
On the host you need:
An NVIDIA GPU and driver. Verify that the GPU is visible with
nvidia-smi.Git LFS installed (
sudo apt-get install git-lfs) — the example datasets are stored with LFS.
Cloning the Repository#
Isaac Lab and Isaac Lab-Arena are nested git submodules, so clone recursively:
git clone --recurse-submodules git@github.com:isaac-sim/AutoData.git
If you already cloned without --recurse-submodules, run:
git submodule update --init --recursive
Then pull the LFS-stored datasets:
git lfs install
git lfs pull
If a dataset remains a small text pointer or cannot be opened as HDF5, see Git LFS Pointer Files.
Recommended Docker Installation#
Docker Prerequisites#
In addition to the common prerequisites, install Docker and the
NVIDIA Container Toolkit
(provides --runtime=nvidia / --gpus all).
The Isaac Sim base image lives on nvcr.io, so log in to NGC before pulling it:
docker login nvcr.io # username: $oauthtoken password: <your NGC API key>
Starting the Container#
From the repo root, build (first run) and enter the base dev container:
./docker/run_docker.sh
The first run builds the default development image, then drops you into a shell inside the container in the mounted repo. This build process may take up to 30 minutes. Subsequent runs reuse the versioned image and are fast.
For SkillGen workflows, use the cuRobo image instead. cuRobo compiles CUDA kernels for
your GPU architecture (auto-detected via nvidia-smi), so this build is slower and kept in
a separate versioned image tag that coexists with the default one:
./docker/run_docker.sh -c
Note
Inside the container the repo is mounted at /workspaces/autodata and python /
pytest are aliased to Isaac Sim’s interpreter (/isaac-sim/python.sh). Unless you chose
the optional conda installation, run commands in these docs from that directory inside the
container.
Useful flags of ./docker/run_docker.sh:
Flag |
Description |
|---|---|
|
Include cuRobo (required for SkillGen). Auto-detects the GPU arch; override with the
|
|
Host datasets directory to mount at |
|
Force rebuild of the image ( |
|
Verbose. |
|
Show all options. |
For a missing Kit window or an X11 error, see Display and X11.
Optional Conda Installation#
Use the conda route if you want to manage the environment and its packages directly. Docker remains the recommended route because it provides the project’s reproducible, preconfigured environment.
The conda installation requires conda and uv on your PATH.
From the repository root, create the autodata environment with Python 3.12:
./conda_installer.sh -c
Activate the environment and install Isaac Sim, CUDA-enabled PyTorch, Isaac Lab, Isaac Lab-Arena, and AutoData:
conda activate autodata
./conda_installer.sh -i
You can create the environment and install the packages in one command:
./conda_installer.sh -c -i
Run ./conda_installer.sh -h to see all installer options. Activate the environment before running
commands from the rest of the documentation:
conda activate autodata
Installing cuRobo for SkillGen#
SkillGen additionally requires cuRobo. Before installing it, review the NVIDIA cuRobo license in curobo-license.txt.
Warning
Install cuRobo from a clean shell that has not sourced Isaac Sim environment scripts such as
setup_conda_env.sh. Those scripts set PYTHONHOME and PYTHONPATH to use Kit’s bundled
packages, which can cause conda to fail during the cuRobo installation.
Activate the AutoData environment, install the CUDA 12.8 toolkit, and configure the build for your GPU’s compute capability:
conda activate autodata
conda install -c nvidia cuda-toolkit=12.8 -y
export CUDA_HOME="$CONDA_PREFIX"
export PATH="$CUDA_HOME/bin:$PATH"
export LD_LIBRARY_PATH="$CUDA_HOME/lib:$LD_LIBRARY_PATH"
export TORCH_CUDA_ARCH_LIST="$(nvidia-smi --query-gpu=compute_cap --format=csv,noheader | head -n1)+PTX"
Install the cuRobo commit tested with Isaac Lab and used by the AutoData cuRobo container:
pip install -e "git+https://github.com/NVlabs/curobo.git@ebb71702f3f70e767f40fd8e050674af0288abe8#egg=nvidia-curobo" \
--no-build-isolation
The editable installation clones cuRobo into src/nvidia-curobo beneath the current directory.
Run the command from the directory where you want to keep that source checkout.
Verify the installation:
python -c "import curobo; print('cuRobo installed successfully')"
For missing modules, CUDA kernel errors, or a GPU change, see cuRobo Build and GPU Compatibility.
Tip
If the import fails because libstdc++.so.6 does not provide GLIBCXX_3.4.30, update the
environment’s C++ runtime libraries:
conda config --env --set channel_priority strict
conda config --env --add channels conda-forge
conda install -y -c conda-forge "libstdcxx-ng>=12" "libgcc-ng>=12"
Verifying the Installation#
Docker users should run these commands inside the container. Conda users should run them from the
repository root after activating the autodata environment.
Run the fast unit tests (a few seconds, no Isaac Sim launch):
pytest autodata_tests -m "not with_subprocess"
Then, to verify the full stack end-to-end, run one data-generation test (launches Isaac Sim as a subprocess; several minutes):
pytest -s autodata_tests/e2e/test_mimicgen_data_generation.py::test_franka_cube_stack_mimicgen_data_generation_single_env_cuda
See Testing and CI for the full test-suite layout.