Ingesting Maps in a CARLA package

This section describes the process of ingesting maps into a package (binary) version of CARLA. If you are using a version of CARLA that has been built from source to ingest maps then follow the guidelines here instead.

This process is only available for Linux systems. The import process involves running a Docker image of Unreal Engine to import the relevant files and then export them as a standalone package which can then be configured to be used in CARLA. The Docker image takes around 4h and 600-700 GB to be built. This is only needed the first time the image is built.


Before you begin

  • You will need to fulfill the following system requirements:
    • 64-bit version of Docker in Ubuntu 16.04+
    • Minimum 8GB of RAM
    • Minimum 700 GB available disk space for building container images
    • Git version control
  • Ensure you are using a package (binary) version of CARLA. If you are using a version of CARLA that has been built from source to ingest maps then follow the guidelines here instead.
  • You should have at least two files, <mapName>.xodr and <mapName>.fbx that have been generated from a map editor such as RoadRunner.
  • These files should have the same value for <mapName> in order to be recognised as the same map.

Map ingestion in a CARLA package

1. CARLA provides all the utilities to build Unreal Engine in a Docker image and to compile CARLA using that image. The tools are found in the source code available in GitHub. Clone the repository using the following command:

    git clone https://github.com/carla-simulator/carla

2. Build the Docker image of Unreal Engine by following these instructions.

3. Create an input_folder. This is where you will put the files to be imported. Docker will automatically create a .json file describing the package folder structure. Change permissions on the input_folder for this to be created successfully:

    #Go to the parent folder, where the input folder is contained
    chmod 777 input_folder

Note

This is not necessary if the package is prepared manually, and contains a .json file already.

4. Create an output_folder. This is where the Docker image will write the output files after it has cooked the map.

5. Navigate to ~/carla/Util/Docker. This is where the ingestion script is located. The script requires the path for the input_folder and output_folder and the name of the package to be ingested. If a .json file is provided, the name of that file is the package name, if no .json is provided, the name must be map_package:

    python3 docker_tools.py --input ~/path_to_input_folder --output ~/path_to_output_folder --packages map_package

Warning

If the argument --packages map_package is not provided, the Docker image will make a package of CARLA.

6. The package will be generated in the output_folder as <map_package>.tar.gz. This is the standalone package that is now ready to be imported into CARLA. Move the package to the Import folder in the CARLA root directory (of the package/binary version where you will be using the map), and run the following script from the root directory to import it:

        ./ImportAssets.sh

7. To run a simulation with the new map, run CARLA and then change the map using the config.py file:

    cd PythonAPI/util
    python3 config.py --map <mapName>



Your map is now ready to run simulations in CARLA. If you have any questions about the process then you can ask in the forum or you can try running some of our example scripts on your new map to test it out.