URDF and TF: Frames for a Tracked Base, LiDAR and RGB-D Camera

What this guide proves

The robot has one unambiguous base frame and measured transforms to its LiDAR and optional RGB-D camera. RViz can display the physical arrangement without a guessed axis or duplicated transform publisher.

Hardware and scope boundary

COG-T1-TRACK-BASE is the mobile structure. COG-MS200-LIDAR supplies one horizontal 2D scan plane; COG-AURORA930-PRO supplies aligned depth, RGB and IR streams. COG-RPI5-8GB hosts ROS 2 and COG-RRC-LITE may provide low-level feedback, but neither product defines the final transforms for a custom installation.

Prepare before applying power

Mount sensors rigidly, route cables without crossing the MS200 scan plane, and measure from a reproducible physical reference. Record forward direction, mounting height, yaw orientation and whether the camera optical frame follows its driver convention.

Procedure

1. Choose base_link at a stable, documented chassis reference and define x forward, y left and z up consistently across URDF, controller and navigation parameters.
2. Add fixed transforms only from measured values. Use temporary static transforms for bench validation, then move final values into the robot description or approved launch configuration.
3. Check the LiDAR frame while the robot is still. A known target placed in front of the chassis must appear in the expected forward direction; do not compensate a reversed scan with a map setting.
4. For Aurora930 Pro, inspect RGB, depth, IR and camera optical frames from the delivered driver. Confirm the parent frame and alignment behavior before consuming depth in an application.
5. Visualize the complete TF tree and sensor data in RViz. Restart the launch system once to detect duplicate broadcasters or transforms that exist only in a local shell.

Pass criteria and record

Save the URDF/Xacro source, measured transform table, sensor mounting photos and a TF-tree snapshot. The same result must return after reboot and when the robot is placed in a known orientation.

Do not proceed when

Do not map, navigate or plan arm motion with a guessed sensor axis, conflicting TF publishers, a LiDAR frame that sees forward as another direction, or an unverified camera optical-frame convention.

Continue from here

Feed only this verified geometry into LiDAR SLAM, RGB-D processing and any arm-to-base transform for mobile manipulation.