AprilTag Pose Estimation Benchmarking

note

To learn more about how to use April Tags, see this grasping how to guide or this tutorial Fiducial Marker Detection with AprilTags

This how-to guide will walk through how to run and visualize an AprilTag Benchmarking example to evaluate the accuracy of fiducial pose estimation.

Launch MoveIt Pro

We assume you have already installed MoveIt Pro to the default install location. Launch the application using:

moveit_pro run -c april_tag_sim

Important: AprilTag Configuration in Simulation

When using AprilTags in simulated configurations (such as MuJoCo), it is critical to understand how tag size affects pose estimation accuracy.

AprilTag Texture White Margins

When creating AprilTag textures for simulation, be wary that AprilTag size is measured by the black square pattern only. Passing in the incorrect tag_size will lead to inaccurate pose estimates. If you use an AprilTag texture that includes a white margin, make sure to measure the size of the black pattern.

For example, in the moveit_pro_example_ws, the AprilTag images have a white border that make up 20% of the texture's dimensions, so the expected size of the tag on a square MuJoCo surface is 80% of the square's side length.

One other common mistake is that the MJCF modeling language uses half-sizes for defining some geometric objects. If you are determining the AprilTag size from a MJCF box geom object, make sure to double the size value.

1. Collect AprilTag Estimation Data

Run the Collect AprilTag Detection Data Objective. This Objective will run AprilTag detection on each of the 6 tags in the workspace 20 times: 10 with the wrist camera parallel to the tag and 10 with the wrist camera angled 45 degrees relative to the tag. The deviation of each detection from ground truth will be recorded and saved to YAML files for analysis in the next step.

2. Analyze AprilTag Estimation Data

While Pro is running, enter the container.

moveit_pro shell

Navigate to the april_tag_sim/scripts directory.

cd ~/user_ws/src/april_tag_sim/scripts

Ensure the required Python packages are installed.

pip install -r requirements.txt

Run the analysis script with python.

python3 analyze_apriltag_data.py

This script will read the data recorded in the previous step and generate plots like the following.