Running CARLA in a Docker

This tutorial is designed for:

  • People that want to run CARLA without needing to install all dependencies
  • Recommended solution to run multiple CARLA servers and perform GPU mapping
  • People who don't need to render the full simulation (the server is headless)

This tutorial was tested in Ubuntu 16.04 and using NVIDIA 396.37 drivers. This method requires a version of NVIDIA drivers >=390.

Docker Installation

Note

Docker requires sudo to run. Follow this guide to add users to the docker sudo group https://docs.docker.com/install/linux/linux-postinstall/

Docker CE

For our tests we used the Docker CE version. To install Docker CE we recommend using this tutorial

NVIDIA-Docker2

To install nvidia-docker-2 we recommend using the "Quick Start" section from the nvidia-dockers github.

Getting it Running

Pull the CARLA image.

docker pull carlasim/carla:version

For selecting a version, for instance, version 0.8.2 (stable), do:

docker pull carlasim/carla:0.8.2

Running CARLA under docker:

docker run -p 2000-2002:2000-2002 --runtime=nvidia -e --gpus all carlasim/carla:0.8.4

The -p 2000-2002:2000-2002 argument is to redirect host ports for the docker container. Use --gpus '"device=<gpu_01>,<gpu_02>"' to specify which GPUs should run CARLA. Take a look at this NVIDIA documentation to learn other syntax options.

You can also pass parameters to the CARLA executable. With this you can chose the town and select the port that is going to be used:

docker run -p 2000-2002:2000-2002 --runtime=nvidia -e NVIDIA_VISIBLE_DEVICES=0 carlasim/carla:0.8.4 /bin/bash CarlaUE4.sh  < Your list of parameters >

At the list of parameters do not forget to add -world-port=<port_number> so that CARLA runs on server mode listening to the <port_number>