RUB RoboRacer

ROS 2 software stack for the RoboRacer (F1TENTH-style) car of the AVAI team, Chair of Software Engineering, Ruhr University Bochum. The car is simulated in Ignition Gazebo (Fortress) with the same ROS interfaces as the real car: LiDAR, RealSense RGB-D camera, IMU, and Ackermann drive commands on /drive.

Parts of this repository are adapted from TUD RoboRacer reference material and have been modified for this setup. Where applicable, source files include license headers or reference notes.

---

Repository layout

rrstack/
├── ros_ws/          ROS 2 workspace (build here)
│   └── src/         ROS 2 packages
├── tools/           Helper scripts (e.g. cone-track world generator)
└── README.md        Installation and quick start (this file)

Package	Purpose	Documentation
roboracer_gazebo	Simulation: car model, worlds, ROS ↔ Ignition bridge	README
roboracer_drive_to_cmd_vel	Simulation only: steering-lag compensator + /drive (Ackermann) → /cmd_vel	README
roboracer_wallfollowing	PID wall-following controller	README
roboracer_lidar_centerline	LiDAR centerline extraction and Pure Pursuit	README
roboracer_slam	LiDAR SLAM with slam_toolbox	README
roboracer_perception	YOLO cone detection	README
roboracer_planning	Cone-based Delaunay waypoints + centreline PID / waypoint pure pursuit	README
roboracer_rl	Reinforcement-learning controller (optional)	README

---

Installation

1. ROS 2 Humble

Requires Ubuntu 22.04. Follow the official guide: https://docs.ros.org/en/humble/Installation/Ubuntu-Install-Debians.html

sudo apt update && sudo apt install -y software-properties-common curl
sudo add-apt-repository universe
sudo curl -sSL https://raw.githubusercontent.com/ros/rosdistro/master/ros.key \
    -o /usr/share/keyrings/ros-archive-keyring.gpg
echo "deb [arch=$(dpkg --print-architecture) signed-by=/usr/share/keyrings/ros-archive-keyring.gpg] \
http://packages.ros.org/ros2/ubuntu $(. /etc/os-release && echo $UBUNTU_CODENAME) main" \
    | sudo tee /etc/apt/sources.list.d/ros2.list > /dev/null
sudo apt update
sudo apt install -y ros-humble-desktop

2. Ignition Gazebo (Fortress) and stack dependencies

sudo apt install -y \
    ros-humble-ros-gz \
    ros-humble-robot-state-publisher \
    ros-humble-xacro \
    ros-humble-ackermann-msgs \
    ros-humble-cv-bridge \
    ros-humble-slam-toolbox \
    ros-humble-nav2-map-server \
    ros-humble-teleop-twist-keyboard \
    python3-colcon-common-extensions \
    python3-rosdep \
    python3-pip

Check the simulator (expected: version 6.x, Fortress):

ign gazebo --version

3. Clone and build

git clone https://github.com/isselab/rrstack.git ~/rrstack
cd ~/rrstack/ros_ws

source /opt/ros/humble/setup.bash
sudo rosdep init      # only once per machine; skip if already done
rosdep update
rosdep install --from-paths src --ignore-src -r -y
pip install -r src/roboracer_perception/requirements.txt

colcon build --symlink-install
source install/setup.bash

Always run colcon build from ros_ws/, never from inside src/.

Optional, only for roboracer_rl:

pip install gymnasium stable-baselines3

---

Launch the simulation

Every new terminal:

cd ~/rrstack/ros_ws
source /opt/ros/humble/setup.bash
source install/setup.bash

Start Ignition Gazebo and spawn the car

ros2 launch roboracer_gazebo gazebo.launch.py                  # walls track (default)
ros2 launch roboracer_gazebo gazebo.launch.py world:=cones     # cone track

This starts Ignition Gazebo, loads the world, spawns the car at the origin, and starts the ROS ↔ Ignition bridge.

Argument	Default	Description
world	walls	walls, cones, or a full path to a .world file
gui	true	false runs the simulator headless
x, y, z, yaw	0.0, 0.0, 0.1, 0.0	Spawn pose of the car
world_name	auto	World name for the reset service; only needed for custom worlds

Examples:

ros2 launch roboracer_gazebo gazebo.launch.py gui:=false
ros2 launch roboracer_gazebo gazebo.launch.py x:=1.0 y:=0.5 yaw:=1.57

Check that the car is running

ros2 topic list                      # /scan /odom /imu /tf /clock /cmd_vel /camera/...
ros2 topic hz /scan

Drive the car manually

ros2 run teleop_twist_keyboard teleop_twist_keyboard

On a virtual machine or a system without a GPU, prefix launch commands with LIBGL_ALWAYS_SOFTWARE=1.

---

Drive the car autonomously

Each launch starts Gazebo, the car, the controller and the simulation-only steering-lag compensator. The car starts driving by itself.

Through cones (camera): YOLO → Delaunay → pure pursuit

ros2 launch roboracer_planning cone_drive.launch.py                    # pure pursuit (recommended)
ros2 launch roboracer_planning cone_drive.launch.py controller:=pid    # centreline PID

camera ─► yolo_node ─► delaunay_node ─► waypoint_pursuit_node ─► /drive ─► steering_lag_compensator ─► drive_to_cmd_vel ─► Gazebo

Gazebo result: 1.25 laps without cone contact, 1.03 m/s average (up to 1.5 m/s). Arguments, topics, parameters and the steering-lag compensation: roboracer_planning README. On the real car: ros2 launch roboracer_planning cone_drive_car.launch.py (no Gazebo, RealSense topics), see Running on the real car.

Along walls (LiDAR): PID wall follower

ros2 launch roboracer_wallfollowing wall_follow.launch.py              # left wall
ros2 launch roboracer_wallfollowing wall_follow.launch.py side:=right  # right wall

Gazebo result: 2.2 laps, 1.34 m/s average. Details: roboracer_wallfollowing README.

Controllers, SLAM, and perception are described in the package READMEs linked above.