create_from_scratch.md

Create a robot container from scratch

We have selected to use containers in order to allow you to be as creative as you can be as a competitor. In the example workspace we provide a Dockerfile which should work for most robots. However, you are free to implement a robot container from scratch. This document provides you some hints.

To develop a container from scratch, please consider reading the Dockerfile provided in the example workspace as a starting point. We would encourage you to do so using a Dockerfile, and then to compile your image using docker build . -t robot_workspace from it. This will allow you to more easily reproduce the work. We use ROS Noetic, however using another version of ROS or ROS2 should work as well.

Hints for starting from scratch

• Be aware of the mounts we are going to use in the docker-compose.yml files found in each task. We will be using these mount points (and only these mount points), so make sure you expose those VOLUMEs in your container for us to use. This probably won't look pretty if you're building from scratch, but you will have to deal with it.
• The simulation container runs a rosmaster at http://simulation:11311, as stated in the ENV field in the Dockerfile in the example workspace. You will need to communicate to this port to interact with Gazebo.
• Make sure your CMD is set correctly - we do not plan on overriding this in the docker-compose.yml file we will use in competition mode.
• If you want to use CUDA in your robot container, have a look at the explanation to use a GPU inside your container.

Beyond that, you are free to completely rebuild the entire environment as you see fit.