🚗 Building an Ackermann-Drive Autonomous Vehicle Simulator in ROS & Gazebo
From Industry-Grade ADAS Systems to Open-Source Robotics
By Chris Sunny Thaliyath
Autonomous driving is no longer limited to large automotive OEMs. With open-source tools like ROS and Gazebo, anyone can experiment with vehicle control, SLAM, and navigation in simulation before touching real hardware.
In this article, I walk through the design of my Ackermann-drive ROS robot simulation and show how the project draws from my real-world experience building a fully autonomous golfcart at Bosch Global Software Technologies (2018–2021).
Repository link:
👉 https://github.com/chrissunny94/ackerman_ros_robot_gazebo_simulation
🏁 Why I Built This Simulation
Vehicles like cars, shuttles, and golfcarts use Ackermann steering geometry—the front wheels rotate at different angles so the vehicle can follow curved paths naturally.
But most robotics tutorials use differential-drive robots, which behave very differently from real cars.
To bridge that gap, I created a simulation environment that behaves like a real autonomous vehicle, complete with:
- Steering linkage geometry
- Physics-accurate wheel dynamics
- LiDAR + IMU sensors
- ROS navigation interfaces
- Ackermann command support
This lets you develop algorithms that transfer more easily to real automotive platforms.
🧩 Project Overview
The repository includes:
✔ Ackermann URDF/SDF Model
Realistic steering joints, wheelbase parameters, and linkages.
✔ ROS Control Integration
Position + velocity controllers for steering and drive motors.
✔ Sensor Simulation
LiDAR, IMU, camera, and optional GPS.
✔ Teleoperation Tools
Drive using keyboard or Ackermann commands.
✔ Launch Files for Gazebo + RViz
One command to bring everything up.
This makes it a powerful environment for:
- SLAM
- Localization
- Obstacle avoidance
- Path planning
- Control tuning
- ADAS prototyping
⚙️ System Architecture
ackerman_ros_robot_gazebo_simulation/
│
├── urdf/ # Complete Ackermann model
├── gazebo/ # Plugins and physics configs
├── config/ # Controller YAML files
├── launch/ # RViz + Gazebo launch scripts
└── scripts/ # Teleop + utilities
The structure mirrors real-world robotic systems, making it easy to extend.
🚀 Getting Started
Clone & Build
git clone https://github.com/chrissunny94/ackerman_ros_robot_gazebo_simulation
catkin_make
source devel/setup.bash
Launch Gazebo Simulation
roslaunch ackerman_robot_gazebo ackerman_sim.launch
Launch RViz Visualization
roslaunch ackerman_robot_description rviz.launch
Drive With Keyboard
rosrun teleop_twist_keyboard teleop_twist_keyboard.py
Publish Ackermann Commands
rostopic pub /ackermann_cmd ackermann_msgs/AckermannDriveStamped
🔬 Real-World Influence: Autonomous Golfcart @ Bosch (2018–2021)
This simulation was heavily inspired by a project I worked on: a fully autonomous low-speed golfcart for internal R&D.
Below is a summary of the core modules I built at Bosch.
📺 Watch a Demo
🛰️ Perception & Localization
At Bosch, I developed the perception stack using:
- Velodyne VLP16 LiDAR
- GPS + IMU fusion via Extended Kalman Filter
- NDT & ICP-based relocalization algorithms
- OpenStreetMap lane overlays for contextual navigation
These concepts can be directly tested in Gazebo by attaching simulated LiDAR + IMU to the robot.
🧭 Planning & Control
I helped build:
- A waypoint follower (contributed upstream to the open-source follow_waypoints repo)
- Lane-Keep Assist (LKA)
- Cruise Control + speed profiling
- Hybrid control loops for Ackermann vehicles
In the simulation, you can replicate these behaviors using:
- Pure Pursuit
- MPC planners
- TEB local planner
- DWB controllers
🎥 Video Demo
🔌 Hardware Integration
On the real golfcart, I built ROS interfaces for:
- CAN bus & DBC decoding
- EPS unit (steering)
- Throttle & brake actuators
- Radar + ultrasonic sensors
- NVIDIA Drive PX GPU acceleration
Even though Gazebo is virtual, these concepts stay relevant because the same ROS interfaces are used to transition from simulation → real hardware.
⭐ Project Milestone: Fully Autonomous Demo
Our team successfully deployed:
- Full LiDAR SLAM
- Live localization + tracking
- Waypoint navigation
- LKA + ACC ADAS features
- Real-time embedded control
This simulation project is my way of bringing that experience back into the open-source world so others can learn and build from it.
🎯 What You Can Do With This Project
Here are practical ideas you can experiment with:
▪ Implement Pure Pursuit or Stanley Controller
▪ Run SLAM (Hector, GMapping, Cartographer)
▪ Test LiDAR-based collision avoidance
▪ Build a modular autonomous driving stack
▪ Port your algorithms to a real robot afterward
The simulation provides the perfect “safe playground.”
🔚 Final Thoughts
Autonomous driving can feel overwhelming, but simulation reduces friction and accelerates innovation.
By blending open-source robotics with real industry practices, this project gives students, researchers, and professionals a practical way to learn, experiment, and build.
Feel free to explore the repository, fork it, and extend it.
👉 https://github.com/chrissunny94/ackerman_ros_robot_gazebo_simulation