murauto/mur_dev_stack repository overview

⁠MUR Docker

Repo for MUR auto's dev stack docker image, refer to the usage section if not trying to compile the docker image.

⁠Features

Base image: nvidia/cudagl:10.0-devel-ubuntu18.04

• CUDA 10.0
• OpenGL (glvnd 1.2)
• cuDNN 7.6.5
• TensorRT 7.0
• ROS Melodic
• Gazebo 9.15.0
• CMake 3.18.2
• OpenCV 4.1.1 (With CUDA/cuDNN)
  • Modify Installers/CMake/opencv_build.sh to change version/compile flags
• Pylon 6.1.1
  • Coded as wget in Dockerfile.
• tkDNN v0.5

⁠Build instructions

Retrive the missing binaries and place it in Installers as such,

Installers/
├─ OpenCV/
│  └─ opencv_build.sh
├─ tkDNN/
│  └─ install_tkDNN.sh
└─ readme.md

Run sudo make to build

⁠Modifying versions

The following docker build args⁠ are exposed with their default values,

ARG CUDA_VERSION=cuda10.0
ARG CUDNN_VERSION=7.6.5.32
ARG TENSORRT_VERSION=7.0.0

ARG CMAKE_VERSION=3.18.2

If the major versions of CUDA/cuDNN/TensorRT changes, it is not enough to simply change the corrosponding ARGs, but modification of the docker file is required, eg. upgrading Tensor RT from 7 to 8 requires libnvinfer7 -> libnvinfer8.

⁠Usage

The docker image is based off of Nvidia's cudagl images, development with CUDA will require an Nvidia GPU, the lastest GPU driver and the Nvidia docker toolkit⁠ on the host machine. Machines without an Nvidia GPU should still be able to run and develop non CUDA based code.

A compiled docker image is hosted on docker hub under murauto/mur_dev_stack⁠.

⁠Simplified Docker launcher script

Refer to https://github.com/MURDriverless/mur_docker_launcher⁠, for a simplified launcher script to automate launching of the docker images.

⁠Manual Method

First allow X clients from any hosts to connect with,

xhost +

^{This allows any one to connect your host machine's X11 server, technically a sercurity fault}

Then launch the docker image with sudo ./run.sh

docker run --gpus all \
    -v /tmp/.X11-unix/:/tmp/.X11-unix/:ro \
    -e DISPLAY=$(echo $DISPLAY) \
    --device=/dev/bus \
    -it \
    murauto/mur_dev_stack

⁠Meaning of the flags

• --gpus all enable pass through of all physical gpus
• -v /tmp/.X11-unix/:/tmp/.X11-unix/:ro mount host's /tmp/.X11-unix/ to the image's /tmp/.X11-unix/ with read-only (:ro)
• -e DISPLAY=$(echo $DISPLAY) export the DISPLAY env variable in the image to point to the host's display
• --device=/dev/bus allows the container access to /dev/bus (USB/Serial device access)
• -it Enable interactive terminal

⁠TODO

• Integrate with Docker hub's CI?
• Find a way to share compiled image (Docker HUB?)
  • Using Docker hub
• Put OpenCV in (Test if opencv is installed properly)
  • Opencv Tested
  • Integrate into dockerfile?
• Integrate tkdnn into dockerfile?
• Split up Installers folder
• Refer to Simplified launcher script⁠
  • Use XAuth alternative to xhost
  • VSCode integration tutorial
  • Exposing Ports and forwarding ros packets to remote machines
• Missing Utilities
  • ifconfig
  • ping
  • vim
• Install gtk and gtk3 module for lidar_dev sudo apt install libcanberra-gtk-module libcanberra-gtk3-module