master-thesis

This repositoriy contains the files and software used in the master's thesis project of Tristan Schwörer and Jonathan Eichild Schmidt in Robotics (MSc) at Aalborg University. The project were conducted during the spring semester of 2023.

Docker

Docker install

https://docs.docker.com/engine/install/ubuntu/

Post install instructions

https://docs.docker.com/engine/install/linux-postinstall/

Docker setting on jetson (ssh into it):

Move docker root to mounted SD card using:

  sudo systemctl stop docker
  sudo systemctl stop docker.socket
  sudo systemctl stop containerd
  sudo mkdir -p /mnt/SDcard/docker/root
  sudo mv /var/lib/docker /mnt/SDcard/docker/root

docker /etc/docker/daemon.json

  {
      "runtimes": {
          "nvidia": {
              "path": "nvidia-container-runtime",
              "runtimeArgs": []
          }
      },
      "default-runtime": "nvidia",
      "data-root": "/mnt/SDcard/docker/root"
  }

docker config is set to the one included in this repository using:

  docker --config /mnt/SDcard/master-thesis/ ps

To make it persistent

  echo export DOCKER_CONFIG=/mnt/SDcard/master-thesis/.docker > ~/.profile

Restart docker for good measure using:

  sudo systemctl restart docker

Check if root directory is correct:

  docker info -f '{{ .DockerRootDir}}'

should return /mnt/SDcard/docker/root

Pull the repository with all submodules using:

git clone --recursive -j8 [email protected]:Tristan9497/master-thesis.git

The docker containers are build on an SDcard that is permanently mounted to /mnt/Sdcard/ to make it viable to deploy on a Jetson. For different paths the docker-compose.yml needs to be adjusted accordingly

Build

Build the two images for the individual Docker containers using (takes hours):

cd /mnt/SDcard/master-thesis/jetson-containers

./scripts/docker_build_ros.sh --distro humble --package ros_base --with-pytorch

cd ..

docker compose -f /realsense-ros2-docker/docker/docker-compose-nvidia.yml build

After build update the image id of the rostorch container in the docker-compose.yml file

Use the docker containers

To use the docker containers for this project the docker-compose.yml file contains the needed instructions and the containers can be run using:

cd /mnt/SDcard/master-thesis
docker compose up

Create a terminal in the rostorch container using:

docker exec -it master-thesis-rostorch-1 bash

Build the ros2 workspace when in rostorch container using:

cd docker-volume/ros_ws
colcon build
source install setup.bash

Run pose_esitmator when in rostorch container using:

ros2 run multi_person_tracker multi_person_tracker

PoseNet

As the roject uses PoseNet from the jetson-inference package, it is important that the network is downloaded before use. This can be done using the tool provided by Nvidia jetson-inference/tools/download-models.sh and downloading Pose-ResNet18-Body

Running simulation with ros1_bridge

The comparison used in this project runs in ROS melodic while the navigation stack is developed for ROS2 Humble. Therefore, a ros1_bridge is used with parameteres. These are found in ros2_ws. The ros_bridge.sh script will source and load all necessary files before running the bridge but must be configured to the setup of the used machine.

Issues

in case of docker error like:

  ERROR [internal] load metadata for nvcr.io/nvidia/l4t-jetpack:r35.2.1 

The docker image needs to be pulled manually using e.g.:

  docker pull nvcr.io/nvidia/l4t-jetpack:r35.2.1 

this was happening when rebuilding the containers

in case of docker pull access denied on docker compose up

 sudo systemctl daemon-reload

 sudo systemctl restart docker