ZCU104 Breakout

AutonomROS: Autonomous Car Swarm with Hardware-accelerated ROS Programming

Paper @ IRC-2023:

Paper: AutonomROS: A ReconROS-based Autonomonous Driving Unit

We present an Ackermann-based autonomous driving robot car based on an ROS 2 architecture alongside accelerated computationally intensive components using hardware acceleration through ReconROS on an Zynq UltraScale+ MPSoC ZCU104 FPGA board. First, our robot car can drive in a lane at a predefined speed. Second, it can detect obstacles and avoid collisions by stopping. Third, our car prevents intersection collisions with other vehicles by communicating via Wifi to a virtual traffic light offering smooth intersection coordination. Finally, since our developed architecture builds upon Navigation 2 (Nav2), a commonly used ROS 2 framework, we provide smooth integration and extendability. Furthermore, we accelerated three computationally intensive architecture components, lane detection, point cloud calculation, and obstacle detection, to save CPU time. In addition, we discovered that leveraging shared memory through Iceoryx as the ROS 2 middleware further reduced CPU time. Hence, we extended ReconROS’s capabilities by making it compatible with Iceoryx.

Demo Video

Hardware

• Zynq UltraScale+ MPSoC ZCU104
• Traxxas Racecar Slash 2WD
• Ultrasonic Sensor (HC-SR04)
• Hall Sensor (KY-024)
• IMU (BNO055)
• PMOD0 Breakout Board
• 2D-Camera (SeeCam Cu30)
• 3D-Camera (Astra Pro)
• Wifi Sticks (TP Link TL-WN823NV3)
• Powered USB Hub (CSL USB Hub 3.2 Gen1)
• MakerBeams

Tested on

• Ubuntu 20.04 LTS (Focal Fossa)
• ROS 2 Galactic
• Vivado 2021.2
• amd64 architecture (PC)
• arm64 architecture (ZCU104)

For more detailed information regarding dependencies consider the ansible folder

Installation

0. Install git

If git is not already installed on your computer, install it with:

sudo apt install git

1. Clone the repository

git clone git@git.uni-paderborn.de:clienen/autonomros.git

2. Initialize the submodules

Change into the root of the repository and run the following command:

git submodule update --init

3. Initialize environment

We provide a recommended way to install all requirements using Ansible, which might cause changes on your system. Otherwise look into the ansible playbooks and follow the steps manually.

3.1 Install Ansible

You can either run our script first_install.sh or look into the documentation of Ansible. !We do not provide any warranty for potential damages!

ansible/scripts/first_install.sh

3.2 Run Ansible

source ~/venvs/ansible/bin/activate
cd ansible/
ansible-playbook -i hosts.yml lab.yml -K --extra-vars "main_user_name=YOUR_USERNAME"

Exchange YOUR_USERNAME with your local username. It is required that you have root access via sudo. The BECOME password is your password. You may get an error for the reconros : Build docker image step. If this is the case, restart your computer and run the ansible playbook again.

3.3 Install Vivado

Go to Xilinx website and download Xilinx Unified Installer 2021.2: Linux Self Extracting Web Installer. Move the file into ~/ansible-files/. Then run the following commands:

cd ~/ansible-files/
chmod u+x ./Xilinx_Unified_2021.2_1021_0703_Lin64.bin
./Xilinx_Unified_2021.2_1021_0703_Lin64.bin -- -b AuthTokenGen

Here, you will be asked for your Xilinx Account Credentials. Next, you need to accept the Xilinx EULA and 3rd Party EULA. The following command accepts the EULAs, downloads, and installs Vivado (this may take some hours):

~/ansible-files/Xilinx_Unified_2021.2_1021_0703_Lin64.bin -- -b Install -c ~/ansible-files/install_config.txt -a XilinxEULA,3rdPartyEULA

3.4 Setup Virtual Traffic Light i.e. Installing mosquitto

To install the Mosquitto broker and the Mosquitto command line tools, run:

sudo apt-get install mosquitto mosquitto-clients -y

The Mosquitto broker starts automatically after installing.

Run the following line to start the Mosquitto broker on each system start:

sudo systemctl enable mosquitto

You need to allow anonymous acces by editing /etc/mosquitto/mosquitto.conf and adding:

allow_anonymous true

Be aware of the security implications!

The mosquitto server must be accessible by the board. You can listen on all interfaces by adding the following to the configuration.

listener 1883 0.0.0.0

Be aware of the security implications!

3.5 Board Setup

You first need to setup an SD card. For this follow the steps in board/README.md.

3.6 Computer Setup

Append the following lines to your .bashrc:

source /opt/ros/galactic/setup.bash
export ROS_DOMAIN_ID=53

Reopen the terminal.

Usage

1.1 Building the ReconROS project

Before running the commands below switch to the correct folder and source the ReconROS environment.

cd code/reconros_ws/carcontrol
source ../ReconROS/tools/settings.sh

Build the messages (Only necessary in the first setup)

make build-msg

Build the hardware. Make sure you have enough RAM or swap space (at least 32 GB combined are recommended).

make build-hw

Build the software

make build-sw

Copy the necessary files to the board

make copy

Note: Make sure to create an entry in your ssh config ~/.ssh/config called board for the Zynq UltraScale+ MPSoC ZCU104

Host board
        User xilinx
        HostName BOARD_IP_ADDRESS

1.2 Building the ROS 2 packages

If you have previously performed the steps in the previous section, reopen the terminal to have a clean environment again.

cd code/ros_ws

Update rosdep with:

rosdep update --include-eol-distros

Install the required dependencies:

make dependencies

Build the ROS 2 packages for the board:

make build-arm64

Copy the necessary files to the board:

make board-copy

Update rosdep on the board:

ssh board rosdep update --include-eol-distros

Install the required dependencies on the board:

make board-dependencies

Copy the necessary scripts to the board:

cd code/scripts
make board-copy-scripts

2. Run with rviz2

Note: If anything crashes (except the set_camera_config.sh) rerun the step.

2.1 Setup the rviz2 config

It is advised to use the config for rviz2 provided in code/ros_ws/src/nav2_local/rviz/config.rviz

Start rviz2 with

rviz2 -d code/ros_ws/src/nav2_local/rviz/config.rviz

2.2 Connect to the board and start Roudi

Connect to the board

ssh board 

Run Roudi

./roudi.sh

2.3 In a new terminal: Load the bitstream onto the FPGA and run the delegate threads

Connect to the board

ssh board 

Go in the directory carcontrol

cd carcontrol

Load the bitstream onto the FPGA

./load_fpga.sh

Run the delegate threads

sudo ./run_software.sh

2.4 Start MQTT server for the virtual traffic light

Make sure the MQTT server is running on the computer:

service mosquitto status

2.5 In a new terminal: Start Nav2

Connect to the board

ssh board

Run Nav2

sudo ./start_nav2.sh

Important: Wait until the terminal stops outputting log messages

2.6 In a new terminal: Configure the 2D camera

Connect to the board

ssh board

Run the configuration file

./set_camera_config.sh

Wait approximately 5 seconds and rerun the command:

./set_camera_config.sh

Go back to rviz2 on your local machine and press "Startup" and wait until the terminal running Nav2 stops outputting log messages again. In rvz2 the car and the costmap should be visible at this point.

2.7 In a new terminal: Start the 3D camera

Connect to the board

ssh board

Activate the 3D camera

./activate.sh

Finally you can set a goal pose in rviz2 and the car should start driving towards the goal.

3. Run with Wifi

Note: If you experience issues running ROS 2 over WiFi. Please consider the following steps:

• Often you cannot use rviz2 via WiFi
• Start Nav2 with start_nav2_wifi.sh instead of start_nav2.sh to automatically startup Nav2
• Specify goal poses through the topic /goal_pose instead of using rviz2 GUI.

Contributors

All members contributed equally. Hence, the members are listed in alphabetical order.

Acknowledgment

We especially thank Prof. Dr. Marco Platzner and Christian Lienen for allowing us to work on an exciting project group involving autonomous driving and hardware acceleration. Moreover, we are thankful for the excellent working conditions in the lab provided to us. We would like to thank Christian Lienen for his continuous support and trust through-out the project. Furthermore, we would also like to thank Andre Diekwisch and Torben Gotowik, who have been available to support us with any administrative and equipment needs. Also, we would like to thank Lennart Clausing and Felix Jentzsch for their guidance and support regarding hardware-related issues. Additionally, we thank Elisabeth Schlatt for her assistance with any administrative questions we had. Furthermore, we would like to sincerely thank Prof. Dr. Marco Platzner and the entire working group for giving us the opportunity to participate in this year’s seminar-workshop. It has been a very enjoyable and inspiring experience. Finally, we would like to thank the entire working group and all those who have helped us in any way to bring the project forward. Your support and contributions were valuable to the success of our project.

License

For open source projects, say how it is licensed.

Project status

This project was actively developed and maintained from April 2022 to March 2023. Development has since been terminated, but we will strive to provide support and address any issues that may arise in the future.

If you encounter any bugs or issues with the project, please don't hesitate to file an issue on our GitHub repository. While we may not be actively developing new features for the project, we will do our best to provide timely assistance and bug fixes.