ROS Development Setup

This is a description how to setup my ROS environment. I make some assumptions like shell etc. based on my needs. Don't just blindly copy paste the commands if you don't have the same environment.

Python 3.12 now wants a virtual environment before installing anything with pip. I recommend using venv.

cd $HOME
python3 -m venv python3_env
source $HOME/python3_env/bin/activate # to .zshrc

When installing system tools, venv may not be enough. We need to force:

pip install --break-system-package <your-package>

Setup via Distrobox [Linux only]

These versions are available. Change the commands depending on your needed ROS version.

• ROS2 Jazzy
• ROS1 Melodic

Install distrobox and podman. Example with Ubuntu >=22.10.

sudo apt install distrobox podman qemu-system

Init podman

podman machine init
podman machine start

If you get an error that gvproxy is not installed, do it with

curl -L 'https://github.com/containers/gvisor-tap-vsock/releases/download/v0.8.1/gvproxy-linux-amd64' -o gvproxy
chmod +x gvproxy
sudo mv gvproxy /usr/libexec/podman/gvproxy

Create the environment container. Add --nvidia if you have an nvidia graphics card.

distrobox create --image ubuntu:24.04 --name jazzy

Enter and run the install script for the environment. If my dev setup is used, you can just type init_jazzy_image.sh anywhere.

distrobox enter jazzy
curl -s https://raw.githubusercontent.com/stelzo/dev/refs/heads/main/bin/init_jazzy_image.sh | bash

Exit the container with exit.

In the future you should enter the container like this. You may use an alias for that.

distrobox enter jazzy -- zsh -c "source /opt/ros/jazzy/setup.zsh && exec zsh"

Note

The upper command is already shortened in my dev environment and can be called by just typing jazzy anywhere.

For Rust, follow the instructions in this section. The script already installs the toolchains.

My Best practices

• mkdir -p ~/projects/ros2_ws/src/thirds Third-party ROS2 packages like drivers, etc. Build them from source where possible. If building is too complicated, look for alternatives or build them yourself - bloat is bad, hiding it is worse.
• mkdir -p ~/projects/ros2_ws/src All ROS2 packages that I develop.
• mkdir -p ~/projects/thirds Third-party libraries from source, Pull Requests, etc.
• mkdir -p ~/projects ROS-independent personal projects, experiments and libraries.
• Create complex logic as libraries and use ROS only as a communication layer or frontend. State Management Nodes like State Machines or Behavior Trees are the exception.
• Configuration is a logic-level concern. ROS has APIs to get them, but you give it via parameters to the library (e.g. structs).
• Remove dependencies as much as possible without adding a huge maintenance burden. Copy code if only a small part is needed and avoid large libraries in general. Any step beyond git clone should be avoided or automated.

Messages

... are language independent descriptions like meta-packages. I use CMake for them, since C++ toolchain is installed with ROS anyways and people may want to use the message but not the Rust package.

Messages use their own repository. This way they can be shared easily with other nodes via a simple git clone. Here is an example repository used in one of my Rust package.

Managing this setup

Export a new image with

podman container commit -p jazzy <your-image-name>
podman save <your-image-name>:latest | bzip2 > <your-image-name>.tar.bz

ROS1 -> ROS2

For comparing old vs new methods, the datasets should be the same. Bagfiles can be converted with

pip3 install rosbags --break-system-packages
rosbags-convert --dst-storage mcap --src <bagfile>.bag --dst <outputfolder>