• CUDA support via NVIDIA Container Toolkit
• Multi architectures: l4t/arm64, linux/amd64
• Multi OS versions:
  • Jetpack: r36.4.0, r35.4.1
  • Ubuntu: jammy, focal
• Multi ros2 distros: humble, galactic, (archived: foxy)
• Multi ros2 packages: core, base, desktop
• Multi packages (non-exhaustive list):
  • Deep-learning: pytorch, torch2trt, tensordict
  • Deep-rl: torchrl
  • Numerical computing: numpy, numba, pycuda, tensorrt
  • MLOps: ray, omegaconf, hydra-core

1. custom dependency management;
2. development environment consistency;
3. codebase stabilization trough continuous integration pipeline;
4. easy deployment to robots compute box;
5. results reproducibility.

Recommanded reading by Tobit Flatscher

• Why should I use Docker when developing robotics software?
• Why is Docker important in particular for academic and research institutions?

Use Dockerized-NorLab project application (DNA), its made to work with DN images.

Note: Dockerized-Norlab images are intended to be used as base images for Dockerized-NorLab project application (DNA).

1. Pick an images that fit your project, the most basic one is dockerized-norlab-dependencies-full

2. Pick the tag that fit your needs, e.g. DN-bleeding-foxy-base-l4t-pytorch-r35.4.1
   Notes: bleeding is for the DN dev branch, foxy is the ros version, base is the ROS package type, l4t-pytorch is the Jetson container pkg, r35.4.1 is the Jetpack version

3. Pull the image from Dockerhub

   docker pull norlabulaval/dockerized-norlab-dependencies-full:DN-bleeding-foxy-base-l4t-pytorch-r35.4.1

4. Pull the image from Dockerhub

    docker run -it --rm \
        --env ROS_DOMAIN_ID=1 --env DISPLAY="${DISPLAY}" --env QT_X11_NO_MITSHM=1 \
        --net host -p 2222:2222 -p 6006:6006 -p 7777:7777 \
        --privileged \
        -v /etc/localtime:/etc/localtime:ro -v /tmp/.X11-unix:/tmp/.X11-unix:rw -v /tmp/.docker.xauth:/tmp/.docker.xauth:rw \
        norlabulaval/dockerized-norlab-dependencies-full:DN-bleeding-foxy-base-l4t-pytorch-r35.4.1

   Note:

   • Add the --runtime nvidia flag if you have nvidia gpu and NVIDIA Container Toolkit is installed.
   • For MacOs user, replace the line --net host -p 2222:2222 -p 6006:6006 -p 7777:7777 with --net bridge.

Notes:

• All images are multi-architecture build (amr64 and amd64).
• Most have CUDA support via NVIDIA Container Toolkit base image, both for L4T (aka Jetson OS) and ubuntu.
• Bleeding tags are build from the latest dev branch version (e.g. DN-bleeding-foxy-base-l4t-pytorch-r35.4.1) and are rebuild each week.
• All images also have a tag pinned for each release version, starting from release v0.5.2, e.g. DN-v0.5.2-foxy-ros-core-l4t-r35.2.1.

 git clone --recurse-submodule https://github.com/norlab-ulaval/dockerized-norlab.git

Assuming a docker builder with multi-architecture docker-container driver named local-builder-multiarch-virtual execute the following in repository root

export BUILDX_BUILDER=local-builder-multiarch-virtual \
  && export NBS_OVERRIDE_BUILD_MATRIX_MAIN=".env.build_matrix.main" \
  && export NBS_OVERRIDE_ADD_DOCKER_CMD_AND_FLAG="build --push" \
  && source dockerized-norlab-scripts/build_script/dn_build_all.bash

Note: To create a multi-architecture docker builder with architecture virtualization

docker buildx create \
    --name local-builder-multiarch-virtual \
    --driver docker-container \
    --node local \
    --platform linux/amd64,linux/arm64 \
    --bootstrap \
    --use