sunsided/ros-gazebo-gpu-docker

Dockerized ROS + Gazebo with GPU support

ROS Kinetic Desktop Full + NVIDIA in Docker

TL;DR: On Linux, to drop into an X11 enabled container created from sunside/ros-gazebo-gpu:kinetic-nvidia, run

./run-nvidia.sh

You'll be user ros with password ros and sudo powers. The current directory will be mounted to /workspace in the container.

---

I am aware that ROS Kinetic is somewhat outdated. However, for some projects I did I had the constraint to work with Kinetic and Gazebo 7 rather than any newer version. Since I was using Ubuntu 20.04 at the time (and ROS Noetic wasn't even released), I created this repo to help me and possibly others out. It is, after all, just a digest of information that can also be found all around the web.

Sourced documentation

• Ubuntu install of ROS Kinetic
• Hardware Acceleration
• NVIDIA Container Toolkit
• Install Gazebo using Ubuntu Packages
• Gazebo Models are available on Git
• nvidia/cudagl Docker image

Problems to steer around

A core issues that arise when just following the official documentation(s) are these:

• The nvidia-docker2 runtime is now deprecated and has been replaced with nvidia-container-toolkit; as a result, --runtime="nvidia" is not a valid option anymore and --gpus "all" (or similar) must be used instead.
• The visualization tools (such as Gazebo and RViz) require OpenGL to work, which isn't available in the offical ROS Docker images (see here and here).
• Gazebo 7.0 is provided with the official ROS Docker image, but is outdated (see here and here). The Docker image comes with Gazebo 7.x from Gazebosim's package sources.
• X11 authentication may fail due to a subtle bug when passing arguments along.

Build with GPU support

For NVIDIA GPU support, go to the docker/nvidia directory and follow the instructions there.

If in a hurry, just run

docker/nvidia/build.sh

This should provide you with the image sunside/ros-gazebo-gpu:kinetic-nvidia.

Start the container with

./run-nvidia.sh

and you're ready to go. Note that this will create a container named ros, so running the command twice won't work. You can, however, docker exec -it ros bash into the running container as often as you like.