Visual Relocation Markers

1. Course Content

Course Overview

In robot autonomous navigation, LiDAR (Light Detection and Ranging) localization is extremely precise, but in the following typical industrial scenarios, it is prone to "localization degradation and drift":

1. Long straight corridors and symmetric shelf passages: Both sides are completely featureless columns with identical characteristics - LiDAR "looks the same everywhere". When the chassis wheels experience slight slipping, the cart has difficulty determining whether it moved forward 1 meter or 1.2 meters;

2. Human traffic and dynamic obstacle occlusion: When LiDAR's field of view is largely blocked by surrounding obstacles, scan matching scores drop sharply;

3. Long-term operation cumulative error: After running for tens of minutes, tiny wheel speed integral errors may accumulate to 5-10 cm, causing the robotic arm to miss objects during grasping operations.

The cart introduces an industrial-grade multi-sensor fusion architecture of "LiDAR localization as primary + visual tag codes (AprilTag/ArUco) global repositioning redundant calibration".

Learning Objectives

1. Deeply understand the geometric calculation principle of AprilTag / ArUco tags in robot visual repositioning.

2. Master the process of graphically checking in and recording tag code global coordinates using apriltag_global_recorder.launch.py tool.

3. Master the data structure of aritag_relocation.yaml configuration file and on-site tag size parameter tuning.

2. Preparation

2.1 Map and Route Network Preparation

Please ensure you have completed the sandbox environment grid map and pose graph construction according to the previous lesson [Build Sand Table Map], and planned the basic driving route network.

2.2 AprilTag Tag Size Confirmation and Parameter Configuration

• The standard sandbox map tag codes included in the kit (usually 36h11 tag family), have a factory-calibrated actual physical black frame outer edge length of 0.048 meters (4.8 cm):

• Core parameter configuration file path:

~/yahboomcar_ros2_ws/dofbot_pro_ws/src/dofbot_pro_bringup/config/common.yaml

• Confirm the actual physical size tag_size of the tag in the parameter file:

xxxxxxxxxx
...
apriltag:                 # Tag detection node configuration
  ros__parameters:
    image_transport: raw  # Image format
    family: 36h11         # Tag family (default is tag36h11)
    size: 0.048           # Tag black outer frame actual physical edge length, unit: meters
...

3. Deploy ArUco Tag Codes in Sandbox Map

• Deployment Principle: Tag codes should be pasted at positions where the onboard camera can clearly see the front or at a large angle when the robot navigates into a workstation (for example, shelf front columns, operation table side panels, or wall edges, recommended height 15-25 cm from ground).

• Recommended sandbox tag deployment reference positions are as follows:

3. Preparation and Execution (Phase 1: Record Tag Code Global Positions)

Before formally using visual repositioning, you need to record and save the global coordinates of the tag codes arranged on the wall or ground in a high-precision LiDAR localization environment.

This phase requires starting the following nodes in sequence:

3.1 Start Chassis and Sensors

Run in Terminal 1 (start chassis drive, LiDAR, and RGB depth camera):

xxxxxxxxxx
ros2 launch yahboomcar_nav laser_bringup_launch.py

3.2 Start RViz Navigation Visualization

Run in Terminal 2 (recommended to open RViz on the virtual machine side when virtual machine and cart are on the same LAN):

xxxxxxxxxx
ros2 launch yahboomcar_nav display_nav_launch.py

3.3 Start Cartographer Pure Localization Node

Load .pbstream pose graph in Terminal 3 to start pure localization core:

xxxxxxxxxx
ros2 launch yahboomcar_nav localization_imu_odom.launch.py use_rviz:=false load_state_filename:=$HOME/map/map.pbstream

3.4 Start Nav2 Navigation Controller

Start navigation controller in Terminal 4 (load corresponding static grid map):

xxxxxxxxxx
ros2 launch yahboomcar_nav navigation_cartodwb_launch.py maps:=$HOME/map/map.yaml

3.5 Start AprilTag Recording Mode

Run in Terminal 5 (start detection node and pose recorder):

xxxxxxxxxx
ros2 launch yahboomcar_nav apriltag_relocation.launch.py mode:=record

1. Click Add in the bottom-left corner of RViz2 interface, select By topic $\rightarrow$ /apriltag/image_rect to add Image plugin for real-time camera detection view:

2. Remotely control the cart to drive to the front of the tag code, ensure the camera image clearly shows the tag code with 3D coordinate axes and green detection box:

   

3.6 View Tag ID and Record Pose

View currently detected tag ID in Terminal 6:

xxxxxxxxxx
ros2 topic echo /detections --once

After confirming the detected ID (for example, ID is 4), call the service to persistently write the tag's global coordinates in the map to the configuration file:

xxxxxxxxxx
ros2 service call /save_april_tag_pose apriltag_localization/srv/SaveAprilTagPose "{tag_id: 4}"