Carla Installation

Testing Environment:
Motherboard: MSI MS-7A63
OS: Linux (Ubuntu 16.04.1)
CPU: Intel(R) Core(TM) i7-7700K CPU @ 4.20GHz; 8 cores
RAM: 32GB
GPU: NVIDIA GeForce GTX 1080 12GB
NVIDIA Driver: Driver v396.54; OpenGL v4.6.0

Prerequisites

The Carla documentation and GitHub are very useful resources for Carla.

Knowlegde prerequisites:
  • Linux
  • Python
  • Git
System prerequisites:
  • Ubuntu 16.04.1
  • Python 3 (3.5.2)

Note

Although Carla can be run on Windows machines, it is recommened to use a Linux environment for machine learning research.

There are three ways to install CARLA depending on your needs.

  1. Pre-compiled version:
    • + Quick to setup
    • + Small directory size (~5GB)
    • - Cannot edit the default maps
  2. Source version:
    • + Have complete control over simulator environment
    • - Requires separate Unreal Engine 4 installation
    • - Slow to setup
    • - Large file size (~30GB)
  3. Docker container
    • + Very portable
    • + Can be used to run multiple instances of Carla on multiple GPUs
    • - May be challenging for first time docker users

Installation

Note

For all testing purposes we will be using Carla 0.8.2 (stable) version. The Pre-compiled version is recommended for first time users.

Pre-compiled version

  1. Download the [Linux] CARLA_0.8.2.tar.gz compiled version from Carla’s 0.8.2 (stable) Releases GitHub page
  2. Create a new working directory and move the downloaded file there:
mkdir carla
cd carla
mv ~/Downloads/CARLA_0.8.2.tar.gz .
  1. Extract the tar.gz file:
tar -xzvf CARLA_0.8.2.tar.gz

Note

The tar arguements you just used are as follows:
  • x: eXtract
  • z: use gZip uncompression algorithm
  • v: provide Verbose output
  • f: next arguement would be the tar.gz File

  1. Run Carla:

    ./CarlaUE4.sh

This should open the Carla environment where the following keys can be used to control the vehicle:

Manual Control:
W - accelerate
A - turn left
D - turn right
S - deccelerate
Q - switch gears (reverse/forward)
P - autopilot

Source version

1. Follow the instructions in Carla documentation’s ‘How to build on Linux’ page. Be sure to install Carla 0.8.2 (stable) source code version alongside Unreal Engine 4.18.

Note

To keep clean directories, it may help to install Unreal Engine in the ~\packages\UnrealEngine_4.28 directory and carla in the ~\projects\carla directory.

Note

If you have previous Anaconda installations, you may need to comment out the lines pertaining to them in the ~/.bashrc file.

Note

If make commands do not work for building carla, run the Setup.sh file. This would involve a ~10GB asset download.

Note

If you run into an issue while running the above commands, with errors like lib.so is missing, you should run Rebuild.sh.

  1. Once both Unreal Engine and Carla have been installed, navigate to the carla folder and run the following commands:
export UE4_ROOT=<your_path_to_UE4>/UnrealEngine_4.18
${UE4_ROOT}/Engine/Binaries/Linux/UE4Editor "<your_path_to_carla>/carla-0.8.2/Unreal/CarlaUE4/CarlaUE4.uproject"

  1. To make your life easier you can put them in a shell script:

    touch CarlaUE4.sh
echo 'export UE4_ROOT=<your_path_to_UE4>/UnrealEngine_4.18' >> CarlaUE4.sh
echo '${UE4_ROOT}/Engine/Binaries/Linux/UE4Editor "<your_path_to_carla>/carla-0.8.2/Unreal/CarlaUE4/CarlaUE4.uproject"' >> CarlaUE4.sh
chmod +X CarlaUE4.sh

Then to run the script you can use the command ./CarlaUE4.sh.

This will open up the UE4 Editor with Carla loaded into it.

  1. Press the play (right point arrow head) above the viewport to run Carla within the Unreal Editor. This will let you control the vehicle.

The following keys can be used to control the vehicle:

Manual Control:
W - accelerate
A - turn left
D - turn right
S - deccelerate
Q - switch gears (reverse/forward)
P - autopilot

Docker version

Docker containers wrap a piece of software in a complete filesystem that contains everything needed to run: code, runtime, system tools, system libraries – anything that can be installed on a server. This guarantees that the software will always run the same, regardless of its environment.

The docker version of CARLA is usually meant to be run on a server, and is typically headless (without a simulator window). This decreases the overhead for rendering the environment. The primary instruction set for installing the CARLA docker can be found in the official documentation.

Note

If you are a Washington University researcher, steps 1 and 2 can be skipped.

  1. Install Docker CE (Community Edition) for your operating system:

2. Install nvidia-docker2 using the official NVIDIA guide.

  1. Pull the CARLA image:
docker pull carlasim/carla:0.8.2

Note

sudo will be required for running docker commands unless you are added to the docker usergroup. Use thisguidehttps://docs.docker.com/install/linux/linux-postinstall/ to do so. If you do not have sudo access on the server that you are using, ask your server admin to add you to the docker group.

Note

The CARLA docker image that is pulled does not contain the PythonClient files. Additionally, it does not contain Python libraries. It is meant to be run as a standalone server to which you can connect a Python client. Use the pre-compiled guide from above to acquire the PythonClient files.

Reference:

Carla Server

When we have been using WASD to control the vehicle in Carla, we have been using Carla as a standalone package. But if we want to it communicate with Python, we need to run it as a server.

Notice the PythonClient\ folder in the carla directory.

Carla has the following parameter arguements:

Pre-compiled version

Note

It may be a good idea to limit the fps and resolution while running carla.

  1. On terminal one, run:

    ./CarlaUE4.sh -carla-server /Game/Maps/Town02 -benchmark -fps=10 -windowed -ResX=800 -ResY=600

  2. Open a new terminal navigate to the PythonClient directory:

    cd PythonClient

  3. Append the carla directory to the Python path, to allow python to use the carla library:

    export PYTHONPATH=~/projects/carla-0.8.2/PythonClient:$PYTHONPATH

To prevent you from entering this command every time, you can add the above line to your ~/.bashrc file and run the command source ~/.bashrc which refreshes your linux profile parameters.

  1. Run one of the Python examples in the PythonClient directory. Example:

    python3 manual_control.py

Note

This may require the pygame library. Install it using the command: pip3 install pygame --user.

This lets you control the vehicle using the WASD keys on the PyGame window.

../../_images/manual_drive.jpg

Carla simulator (top) with PyGame window (bottom) camera view

Note the bottom left mini-image is the depth map and the on the the right is the semantic segmented camera view.

Note

As an alternative to using the arguement -carla-server, the Example.CarlaSettings.ini file in the carla directory can be edited to suit your need. For example UseNetworking=true would run carla as a server without using -carla-server. But you would have to pass that .ini file as a param: ./CarlaUE4.sh -carla-settings=Example.CarlaSettings.ini

Source version

  1. Open the carla/Unrael/CarlaUE4/Config/CarlaSettings.ini file with a text edit.
  2. Change UseNetworking=false to UseNetworking=true
  3. Press the play button, and follow steps 2 to 4 from the above (pre-compiled) section.

Docker version

  1. Run the CARLA server using Docker:
docker run -p 2000-2002:2000-2002 --runtime=nvidia -e NVIDIA_VISIBLE_DEVICES=0 carlasim/carla:0.8.2

Note

-p 2000-2002:2000-2002 argument is to redirect host ports for the docker container.

Note

NVIDIA_VISIBLE_DEVICES=0 makes the docker run on GPU ID 0. GPU listing can be acquired using the command nvidia-smi.

2. Open a new terminal and navigate to the PythonClient directory from the pre-compiled version.

  1. Run one of the Python examples in the PythonClient directory. Example:

    python3 manual_control.py

Note

This may require the pygame library. Install it using the command: pip3 install pygame --user.