Could the carla rllib-integration run in docker?

[ll7]

Hello,

do you think it is possible to run the rllib integration in docker? My idea of the process so far has been:

• creating a Dockerfile
• From carlasim/carla:0.9.11
• exporting CARLA_ROOT
• clone rllib-integration into the container
• installing the requirements and dqn_reqiurements
• enabling a gui? is this necessary or does everything run headless?
  • either using -e SDL_VIDEODRIVER=x11 -e DISPLAY=$DISPLAY -v /tmp/.X11-unix:/tmp/.X11-unix
    • sudo xhost +local:root
  • or using a VNC viewer to provide a display output
• docker build
• docker run with port connection for tensorboard?
  • does pytorch from the dqn example use tensorboard?

Does anyone have experiences with a docker setup?

Kind regards and thanks for sharing the rllib-integration.

[joel-mb]

Hi @ll7 !
I think this is totally doable. Maybe the only missing step is adding the CARLA Python API to the PYTHONPATH.

enabling a gui? is this necessary or does everything run headless?

Actually, there is one parameter in https://github.com/carla-simulator/rllib-integration/blob/main/dqn_example/dqn_config.yaml#L50 to that end. For testing purposes, it may be useful to enable the gui.

docker run with port connection for tensorboard?
does pytorch from the dqn example use tensorboard?

yeah, the dqn_train.py example is automatically opening a tensoboard server in the port 6006.