A 4-wheel differential-drive rover built from scratch as a ROS 2 package: URDF model, Gazebo simulation, a 360° 2D LiDAR and an Intel RealSense D455-style depth camera. It is the simulation platform I used to develop LiDAR- and depth-camera-based autonomous navigation (apf-navigation) before moving to a real rover.
| Part | Details |
|---|---|
| Chassis | 1.0 × 0.6 × 0.3 m box body |
| Drive | 4 continuous wheel joints, 0.3 m diameter, 0.8 m track. Two gazebo_ros_diff_drive plugins (front and rear pairs) give 4-wheel skid-steer behaviour from a single /cmd_vel |
| 2D LiDAR | 360 samples over 360°, 10 Hz, range up to 12 m → /scan (laser_frame) |
| Depth camera | Modelled on the D455: 87° horizontal FOV, 848 × 480, 0.3–6 m, 30 Hz → /camera/depth/image_raw with a proper optical frame (camera_link_optical) |
| Odometry | Wheel odometry and TF from the drive plugins, plus ground-truth pose from gazebo_ros_p3d on /odom |
Two variants:
urdf/four_wheel_diffdrive.urdfis the LiDAR-only rover.urdf/depth_four_wheel_diffdrive.urdfis the LiDAR plus depth camera rover.
Tested on: ROS 2 Humble, Gazebo Classic 11 (gazebo_ros_pkgs).
mkdir -p ~/ws_diffdrive/src && cd ~/ws_diffdrive/src
git clone https://github.com/3AMengineer/four_wheel_diff_drive.git
cd .. && colcon build --packages-select four_wheel_diff_drive
source install/setup.bash
# LiDAR rover
ros2 launch four_wheel_diff_drive gazebo_model.launch.py
# or: LiDAR + depth camera rover
ros2 launch four_wheel_diff_drive depth_gazebo_model.launch.py
# Drive it
ros2 run teleop_twist_keyboard teleop_twist_keyboardIn RViz, set the fixed frame to odom and add /scan, /camera/depth/image_raw and the RobotModel display.
- Frames matter more than geometry. A wrong joint origin or a camera without an optical frame looks fine in Gazebo and silently breaks everything downstream.
- Skid-steer isn't textbook differential drive. Four driven wheels slip when they turn, so the odometry from the plugins drifts. That's why the navigation stack compares it against ground truth.
- Simulating the sensor you'll actually use pays off. Matching the D455's FOV and range early made the move from LiDAR to depth perception much less surprising.
- apf-navigation: artificial potential field obstacle avoidance on 2D LiDAR and on D455 depth images
- The UAS-DTU autonomous rover for the DARPA Triage Challenge: ROS 2, Nav2, a Blickfeld 3D LiDAR, LiDAR-inertial odometry and ArduRover on real hardware
Shourya Sharma, Mathematics & Computing @ DTU, UAS-DTU LinkedIn