Dev Env

A batteries-included opinionated environment for UAS Multirotor PX4/ROS2 project development.

Features

• Includes ROS2 Humble, PX4 16.0, uXRCE-DDS, and other development tools/dependencies
• VSCode configuration pre-configured (generic editor documentation coming soon)
• QGroundControl bundled in the system path with qgc
• OpenCV with gstreamer support
• gstreamer webrtc support

See here for the full walkthrough

Container Layout

/build
├── PX4-Autopilot/  // PX4 repository location
└── lib/            // Extra libraries
/home/uas/workspace // Where this dir is mounted as a volume

Repository Layout

.                   // Mounted to /home/uas/workspace in container
├── .devcontainer/  // Container configuration + setup
├── .vscode/        // VSCode configuration
├── repos/          // Repository lists for vcstool
└── uas_ws/         // ROS2 workspace
    └── src/        // ROS2 packages
        └── ...

The PX4 repository is cloned to /home/uas

Getting Started

OS Support:

• Linux (x86_64): Working, any distro, optimal, beautiful, perfect.
• Windows (x86_64): Working*
• MacOS: Non-working (Gazebo does not work)
• Windows (ARM): untested
• Linux (ARM): untested
• FreeBSD: untested
• Linux (RISC-V): untested

*Gazebo seems unable to use integrated GPUs however, and dedicated GPU setup has not been tested

Recommended at least ~30GB of free space (more is better).

• Container is about ~23GB (it takes a lot of packages to run this stuff).
• Extra 2GB for the container storage, files, caching, logs, etc.
• ~5GB for build artifacts, cache, etc.

Assuming you have Docker and VSCode already installed. (For Windows Docker should be installed via WSL in order for X11 apps to work) (For Linux rootless Docker will not work as the workspace ownership will be weird)

See Virtual Desktop for enabling a browser accessible desktop.

1. Clone this repository and open it in VScode. Ensure you have Dev Containers extension installed.
2. Take the time now to enable GPU passthrough for Gazebo and Computer Vision tasks.
3. You should be prompted by a pop-up in the bottom right to open this folder in a container.
   • If not, open command palette ctrl+shift+p and search for Dev Containers: Open Folder in Container.
4. VScode should now pull the container from ghcr.io and run some setup commands
5. Test that your setup works by running the Launch Simulator task
6. For version control GitHub via SSH is required: GitHub docs
7. Enable ssh-agent passthrough for devcontainers.
8. From /home/uas/workspace Run vcs import < ./repos/latest.repos
   • If new repositories do not appear in VScode source control, from command palette run Developer: Reload Window to refresh.
   • If new repositories are still not visible on the source control tab click the ... then View & Sort > repositories and check the ones you want to see. Alternatively under Views enable Source Control Repositories for an alternate layout.
9. Once in the container run cd uas_ws and run colcon build --symlink-install.
10. Read each repositories README.md for more information on working with them.

vcstool

We use vcstool to easily manage the repositories and checkout to specific commits. All examples are run in ~/workspace (default dir). Aliases have been provided for convenience in the comment #.

For example to fetch (and switch) to the latest commit:

vcs import < ./repos/latest.repos # vcsl

Then assuming you have no conflicting changes, pull the latest commits into your local repo:

vcs pull -n # vcsp

You can also checkout to specific tags or branches, defined in .repos file, for example to use the stable tag:

vcs import < ./repos/stable.repos

Managing remote and local tags are tricky, if you would like to pull (overwrite) tags from remote then:

vcs custom -n --git --args fetch --tags --force # vcst

GPU

A GPU is highly recommend for Computer Vision tasks. It is also nice to have for Gazebo otherwise you will get around 12 fps (although the RTF factor seems fine).

Setup will vary widely between OS and hardware configurations, it is likely best you consult external sources for your specific configuration.

For pytorch change which version gets install in .devcontainer/on-create.sh

Nvidia

Windows/WSL2:

TODO DOCUMENTATION NEEDED

Ubuntu:

TODO DOCUMENTATION NEEDED

NixOS:

hardware.nvidia-container-toolkit.enable = true;

Uncomment the deploy section in docker-compose.yaml

AMD

Pytorch does not work with AMD cards earlier than the RX 6000 (Navi 2x) series

TODO DOCUMENTATION NEEDED

Intel

Pytorch does support Intel GPUs.

TODO DOCUMENTATION NEEDED

iGPU

Linux: If it works on the host it should just work

Windows/WSL2: Seems to be unsupported for Gazebo

Building the image

To build the image from scratch you can run:

docker build dev_env:local .devcontainer

Change the image in ./devcontainer/docker-compose.yml

Virtual Desktop

Warning large performance hit (~2 times slower when run on my machine), use only if necessary.

All files in .devcontainer directory.

In devcontainer.json uncomment the features object and ports 6080, 5901 in forwardPorts and portAttributes.

In docker-compose.yml comment out environment: and associated intended variables.

Rebuild the container (it should hopefully use the cached layers), and open localhost:6080 in your browser, login and you shoudl see a desktop. All GUI applications will now open in here when run from the terminal.

Tips:

• On the pullout tab on the left under settings set Scaling Mode: to Remote Resizing to set the resolution of the desktop to your own.

Customization

If you have a preferred workflow/tools that are not installed/enabled by default (you will know if this is you). Instead of polluting the Dockerfile with random packages/scripts add a custom setup script in .devcontainer/profiles/, then set the USER_PROFILE var in .devcontainers/post-create.sh. Keep in mind this script runs once after container creation. Submit a PR with your setup script if you think its useful/interesting.

Or use an existing devcontainer-feature

Adding VSCode extensions can be done via:

https://code.visualstudio.com/docs/devcontainers/containers#_always-installed-extensions

Installing more software in the container

The container is based on Ubuntu 22.04.5 LTS, the uas user is configured with passwordless sudo.

# Probably have to run this the first time you open the container
sudo apt update
# Install more software e.g. btop
sudo apt install btop
# Use pip (as sudo unless using venv) to install more python modules (only if necessary)
sudo pip install pyjokes

QGroundControl

QGroundControl is bundled with the container and will work with the PX4/Gazebo simulation. Simply run qgc.

Joystick

Enable usb joystick by uncommenting the /dev/input:/dev/input volume in .devcontainer/docker-compose.yml (security note: this passes through all of your input devices to the container). Depending on your joystick you may need https://github.com/sezanzeb/input-remapper/tree/main?tab=readme-ov-file, or edit and export the SDL_GAMECONTROLLERCONFIG environment variable. See ./profiles for examples.

In QGroundControl on the Vehicle Configuration page the Joystick option should be visible, enable joystick input, calibrate if necessary and assign button actions.

Depending on the simulation/Docker/OS/hardware you will get communication loss errors from QGroundControl, set COM_RC_LOSS_T to some large number like 5 seconds, probably don't do this on a real drone but for simulation its fine.

Configuration

If you want to allow an instance of QGroundControl running natively on the host system to connect to the simulation, goto .devcontainer/docker-compose.yml and uncomment network_mode: host, you should also comment out forwardPorts in .devcontainer/devcontainer.json to fix some behavioral issues. Rebuild the container.

After rebuild make sure that no ports are forwarded from the devcontainer, by default the ports tab is on the same panel as the terminal. Select any open ports and right click stop forwarding. (this is the behavior for Linux, other operating systems may be different)

Note that network_mode: host will make it more likely that other issues will occur. You may get firewall issues which will prevent ROS2 from running.