Dockerized-SNOW

^{Project related link:   NLSAR (Youtrack)   •   norlabsnow (Docker Hub)   •   NorLab_MPPI (GitHub)   •   SNOW_AutoRally (GitHub)}  

Containerized development workflow for the NorLab_MPPI and SNOW_AutoRally projects leveraging nvidia-docker technology. Believe it or not, it's configured for developing with ROS melodic in python 3.6.

Key benefit: custom dependency management, development environment consistency, easy deployment to robots compute box and results reproducibility.

Author: Luc Coupal

Container prompt legend

Usage

Quick start for the NorLab_MPPI project on x86 workstation

• Requirement: docker and nvidia container toolkit must be installed (follow install step 1 and 2)
• Tips: You can use the --help flag for usage instruction on most ds_* command

# Create a directory for your development source code if you dont already have one
mkdir -p ~/Repositories && cd ~/Repositories

# Clone both repositories 
sudo git clone https://github.com/norlab-ulaval/NorLab_MPPI.git
sudo git clone https://github.com/RedLeader962/Dockerized-SNOW.git

# Install aliases and check Nvidia NVCC  
cd ~/Repositories/Dockerized-SNOW
source ds_setup.bash

# Pull the norlab-mppi-develop image from norlabsnow Dockerhub with the x86-ubuntu18.04 tag
sudo docker pull norlabsnow/norlab-mppi-develop:x86-ubuntu18.04 

# Create a new docker image instance for development on your machine and start working on the 
# NorLab_MPPI project using ROS melodic, Python 3 and Pytorch right away.
ds_instantiate_develop --runTag=x86-ubuntu18.04 --name=MyCoolName --src="$HOME/Repositories/NorLab_MPPI"

To open an terminal inside MyCoolNmae, use the following convenient script

ds_attach MyCoolNmae

or use sudo docker exec -it MyCoolName bash

Quick start for the NorLab_MPPI project on Apple M1 (arm64) workstation

• It's the same image as the arm64-l4t but with PyTorch and Numba compiled specifically for arm64-Darwin
• Tips: You can use the --help flag for usage instruction on most ds_* command
• Be advise, cuda is not supported on Apple computer so PyTorch and Numba will work on cpu

# Create a directory for your development source code if you dont already have one
mkdir -p ~/Repositories && cd ~/Repositories

# Clone both repositories 
sudo git clone https://github.com/norlab-ulaval/NorLab_MPPI.git
sudo git clone https://github.com/RedLeader962/Dockerized-SNOW.git

# Install aliases
cd ~/Repositories/Dockerized-SNOW
source ds_setup.bash

# Pull the norlab-mppi-develop image from norlabsnow Dockerhub with the arm64-Darwin-ubuntu18.04 tag
sudo docker pull norlabsnow/norlab-mppi-develop:arm64-Darwin-ubuntu18.04

# Create a new docker image instance for development on your machine and start working on the 
# NorLab_MPPI project using ROS melodic, Python 3 and Pytorch right away.
ds_instantiate_develop --runTag=arm64-Darwin-ubuntu18.04 --osx --name=MyCoolName --src="$HOME/Repositories/NorLab_MPPI"

To open an terminal inside MyCoolName, use the following convenient script

ds_attach MyCoolName

or use sudo docker exec -it MyCoolName bash

Quick start for the NorLab_MPPI project on Apple (x86) workstation

• Tips: You can use the --help flag for usage instruction on most ds_* command
• Be advise, cuda is not supported on Apple computer so PyTorch and Numba will work on cpu only

Use a x86 docker image and let docker use roseta to emulate the x86 architecture

# Install roseta
softwareupdate --install-rosetta

# Create a directory for your development source code if you dont already have one
mkdir -p ~/Repositories && cd ~/Repositories

# Clone both repositories 
sudo git clone https://github.com/norlab-ulaval/NorLab_MPPI.git
sudo git clone https://github.com/RedLeader962/Dockerized-SNOW.git

# Install aliases 
cd ~/Repositories/Dockerized-SNOW
source ds_setup.bash

# Pull the norlab-mppi-develop image from norlabsnow Dockerhub with the arm64-Darwin-ubuntu18.04 tag
sudo docker pull norlabsnow/norlab-mppi-develop:x86-ubuntu18.04

# Create a new docker image instance for development on your machine and start working on the 
# NorLab_MPPI project using ROS melodic, Python 3 and Pytorch right away.
ds_instantiate_develop --platform='linux/amd64' --runTag=x86-ubuntu18.04 --osx --name=MyCoolName --src="$HOME/Repositories/NorLab_MPPI"
# Runnning `uname -m` inside the container will confirm the type of architecture

To open an terminal inside MyCoolname, use the following convenient script

ds_attach MyCoolname

or use sudo docker exec -it MyCoolName bash

Quick start for the SNOW_AutoRally project on x86 workstation

• Requirement: docker and nvidia container toolkit must be installed (follow install step 1 and 2)
• Tips: You can use the --help flag for usage instruction on any dockerized-snow bash script

# Create a directory for your development source code if you dont already have one
mkdir -p ~/Repositories && cd ~/Repositories

# Clone both repositories 
sudo git clone https://github.com/RedLeader962/SNOW_AutoRally.git
sudo git clone https://github.com/RedLeader962/Dockerized-SNOW.git
cd ~/Repositories/Dockerized-SNOW

# Pull the norlab-mppi-develop image from norlabsnow Dockerhub with the x86-ubuntu18.04 tag
bash ds_build_dependencies.bash --x86 --GT-AR
bash ds_build_develop.bash --x86 --GT-AR


# Create a new docker image instance for development on your machine and start working on the 
# NorLab_MPPI project using ROS melodic, Python 3 and Pytorch right away.
bash ds_instantiate_develop.bash --runTag=x86-ubuntu18.04 --name=THEgtar --src="$HOME/Repositories/SNOW_AutoRally"



                                          .|'''.|                                               
                                          ||..  '                                               
····································· Dockerized-SNOW ••········································
                                           ''|||.                                               
                                         .     '||                                              
                                         |'....|'                                               

                                 https://norlab.ulaval.ca                                       
                              https://redleader962.github.io                                    

root@norlab-og:/#

Then follow the step at SNOW_AutoRally: Autonomous Driving in Simulation using MPPI

roslaunch autorally_gazebo autoRallyTrackGazeboSim.launch

...

High level architecture

_{(  View diagram fullscreen   •   edit diagram  )}

---

Instruction:

• ★ | Setup PyCharm for local development using a Docker Python interpreter
• Using the nvidia-docker image on a Jetson device (arm64-l4t)
• Using the nvidia-docker image on a x86 host
• Building the arm64-l4t nvidia-docker image on a x86 host using qemu virtualization

How-to

• How-to nvidia-docker manualy (a quick start)
• Test AutoRally Configuration (Revised instruction)
• How-to push image localy builded image to docker hub from command line

Images:

To pull the latest image from docker hub, execute the following in terminal:

sudo docker pull <container name>:<tag>

with <container name>= theImageName and <tag>= theHostArchitecture

---

References:

nvidia-docker Documentation:

• nvidia-docker: Build and run Docker containers leveraging NVIDIA GPUs
  • NVIDIA Container Runtime on Jetson
  • Driver containers
• NVIDIA Cloud Native Technologies
• Base image for jetson:
  • https://ngc.nvidia.com/catalog/containers/nvidia:l4t-base
  • https://developer.nvidia.com/embedded/jetson-cloud-native
• Base image with CUDA and OpenGL support:
  • https://hub.docker.com/r/nvidia/cudagl/
  • https://github.com/NVIDIA/nvidia-docker/wiki/CUDA
  • https://ngc.nvidia.com/catalog/containers/nvidia:cudagl

Docker Documentation:

• Use the Docker command line | Docker Documentation
• Dockerfile reference | Docker Documentation

---