Docker Development

cuRobo will work on most docker images that have pytorch. We provide some example dockerfiles in curobo/docker for reference. The dockerfiles are described in the below table.

Tag	File	Description
x86	x86.dockerfile	Dockerfile that builds cuRobo with a pytorch base container.
aarch64	aarch64.dockerfile	Dockerfile that builds cuRobo with a pytorch base container for using on a NVIDIA Jetson. This container can only be built on a NVIDIA Jetson.
isaac_sim_VERSION	isaac_sim.dockerfile	Dockerfile that builds cuRobo with NVIDIA Isaac Sim VERSION and vulkan. This docker can run Isaac sim with native GUI or headless. Replace VERSION with 2022.2.1 or 2023.1.0.

Building your own docker image with cuRobo

1. Add default nvidia runtime to enable cuda compilation during docker build:

   Edit/create the /etc/docker/daemon.json with content:
       {
       "runtimes": {
           "nvidia": {
               "path": "/usr/bin/nvidia-container-runtime",
               "runtimeArgs": []
           }
       },
       "default-runtime": "nvidia" # ADD this line (the above lines will already exist in your json file)
       }
   

2. If you are building a docker with isaac sim, setup a NGC account following instructions in Isaac Sim Container Setup.

3. bash build_docker.sh TAG, replace TAG with a name from the tag column in the above table.

4. bash start_docker.sh TAG to launch the built docker.

If you want to have a docker that also enables development by mounting a folder, you can create a user docker with the below commands (skip step 4 above). This will mount a folder /home/${USER}/code into the docker container where you can do your development.

4. bash build_dev_docker.sh TAG

5. bash start_dev_docker.sh TAG will start the docker.

Build Warp for NVIDIA Jetson (Deprecated)

Note

Warp is available from pypi starting with 0.11.0. The below instructions are not needed anymore.

NVIDIA Warp requires 11.5+ CUDA to compile and NVIDIA Jetson ships with CUDA 11.4. We will install a new version of CUDA and then compile the library (.so) file for using warp on NVIDIA Jetson.

1. Install a newer version of cuda using the below commands (details):

   wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/arm64/cuda-ubuntu2004.pin
   sudo mv cuda-ubuntu2004.pin /etc/apt/preferences.d/cuda-repository-pin-600
   wget https://developer.download.nvidia.com/compute/cuda/11.8.0/local_installers/cuda-tegra-repo-ubuntu2004-11-8-local_11.8.0-1_arm64.deb
   sudo dpkg -i cuda-tegra-repo-ubuntu2004-11-8-local_11.8.0-1_arm64.deb
   sudo cp /var/cuda-tegra-repo-ubuntu2004-11-8-local/cuda-*-keyring.gpg /usr/share/keyrings/
   sudo apt-get update
   sudo apt-get -y install cuda
   export CUDA_HOME=/usr/local/cuda-11.8/
   

2. Download and compile warp inside a clone of curobo for use inside a dockerfile later:

   git clone https://github.com/NVlabs/curobo.git && cd curobo && mkdir pkgs && cd pkgs
   git clone https://github.com/NVIDIA/warp.git && cd warp && python3 build_lib.py --no_standalone
   

Note

Make sure when running python3 build_lib.py for warp, it’s compiling the CUDA kernels. warp looks for cuda toolkit at CUDA_HOME so set it to your cuda toolkit path before running build_lib.py.