Setup Instructions

The docker setup is only available for computers with NVIDIA-capable GPUs (Minimum 32 GB of free space). TO verify this, run nvidia-smi and verify if you have any problems. If you haven't installed it, go to Software & Updates >> Additional Drivers and install the recommended NVIDIA driver

Verify CUDA on Jetson Nano

cd /usr/bin
sudo ./tegrastats

Create a catkin workspace:

mkdir -p ~/catkin_ws/src && cd ~/catkin_ws
catkin init
cd src
git clone git@github.com:git@github.com:manx52/ROB498.git

Install necessary packages

sudo apt-get install apt-utils python3-pip
pip3 install --upgrade setuptools
sudo apt-get install -y curl

Install docker (https://docs.docker.com/engine/install/ubuntu/)

curl -sSL https://get.docker.com/ | sh

Install docker-compose using pip install method (https://docs.docker.com/compose/install/)

sudo python3 -m pip install docker-compose

Install docker NVIDIA tools

# Make sure you have nvidia drivers (nvidia-smi), if not https://linuxconfig.org/how-to-install-the-nvidia-drivers-on-ubuntu-20-04-focal-fossa-linux
curl -s -L https://nvidia.github.io/nvidia-container-runtime/gpgkey | sudo apt-key add -
distribution=$(. /etc/os-release;echo $ID$VERSION_ID)
curl -s -L https://nvidia.github.io/nvidia-container-runtime/$distribution/nvidia-container-runtime.list | sudo tee /etc/apt/sources.list.d/nvidia-container-runtime.list
sudo apt-get update
sudo apt-get install -y nvidia-container-runtime nvidia-container-toolkit mesa-utils
sudo systemctl restart docker
echo "xhost +local:docker" >> ~/.bashrc && source ~/.bashrc

Running Instructions for Computer

• docker-compose.yaml is for a docker container that runs on a normal computer
• docker-compose.robot.yaml is for a docker container that runs on a Jetson Nano. Building from scratch will take more then an hour.

roslaunch drone mavros_posix_sitl.launch # run simulation
roslaunch drone gui.launch # For visualization

docker-compose -f docker-compose.yaml pull # Use docker-compose build if you want to build locally
docker-compose -f docker-compose.yaml up

Pull and Run the ARM image for Jetson Nano

docker-compose -f docker-compose.robot.yaml pull
docker-compose -f docker-compose.robot.yaml up

Testing Docker for CUDA on Jetson Nano

• Open up 2 terminals
• In one, run docker-compose -f docker-compose.robot.test.yaml up
• In the other run the code below

docker ps # list all of the docker containers that are running
docker exec -it <docker container id> bash
python3
import torch
torch.cuda.is_available()

If the output is true then success CUDA is enabled on the docker.

Controling Drone from remote computer

Make sure your computer and the robot are connected to the same network, then use your personal computer as the ROS_MASTER_URI. You can find the IP address using ifconfig

• On your personal computer (or a dedicated server computer)

export ROS_MASTER_URI=http://<your computer ip>:11311
export ROS_IP=<your computer ip>

• On your person computer edit /etc/hosts and add <robot1_ipaddress> <robot1_hostname> as an a line
• On the robot add this to the bashrc

export ROS_MASTER_URI=http://<your computer ip>:11311

• Verify creating a publisher and subscriber on robot and your computer using rostopic pub, rostopic echo and verify that the nodes can connect to each other using ```rosnode info <subscriber/publisher> node

Reference

Terminology

• image: a file representing an OS
• container: a running instance of an image

Basic docker commands

docker system prune # Clean everything

# Images
docker image ls # list all of the docker images that are built
docker system prune --all # Delete all docker images
docker pull <docker image name> # Pull a docker image
docker tag old_image_name new_image_name
docker run -it <docker image id> bash # Start a docker image as a container

# Containers
docker ps # list all of the docker containers that are running
docker stop $(docker ps -aq) # Stop all running docker containers
docker rm $(docker ps -aq) # Remove all not running docker containers
docker stop $(docker ps -aq) # Stop all not running docker containers
docker exec -it <docker image id> bash # SSH into a docker image to see whats going on inside it

# Identify service with port and kill it
ps -aux | grep <PORT> && sudo kill -9 <PID>
sudo netstat -tulpn | grep <PORT> && sudo kill -9 <PID>

# Docker compose
docker-compose build # Build all docker images
docker-compose pull # Takes images from dockerhub
docker-compose push # Push all docker
docker-compose up # Start the containers specified in this file

Notes

• You never need to build the images a second time as all the code is mounted
• For modifications please refer to the dockerfile and docker-compose.yaml. They are commented and show which sections are important to change
• The initial docker containers are hosted on utrarobosoccer's docker hub here is a link to setup your own. https://docs.docker.com/docker-hub/quickstart/
• if the docker commands give a permission error put a sudo in front of the command