Self-Calibrating Robot Arm Kinematics Correction

ISEF Category: Robotics and Intelligent Machines

Subcategory: Robot Kinematics  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

A tiny angle error at one joint can turn a robot arm's pencil mark into a messy curve. That is why calibration matters. Your phone can help you measure those errors without fancy motion-capture gear. You can turn a cheap arm into a much better one by fixing the math behind it.

What Is It?

This project studies robot arm calibration. Calibration means finding the real values of a robot's geometry, then correcting the model the robot uses to move. A URDF, or Unified Robot Description Format file, is the digital blueprint that tells software where each joint sits and how long each link is.

Think of the arm like a human finger with slightly wrong knuckle positions drawn on a map. If the map is off, every point you try to reach will land in the wrong spot. Denavit-Hartenberg, or DH, parameters are the numbers engineers use to describe each joint's position and twist. Your goal is to compare what the robot should do with what it really does, then use camera observations of a ChArUco board, a board with both checkerboard corners and ArUco markers, to estimate and fix those DH errors.

The cool part is the feedback loop. The camera sees the arm, the software estimates the mismatch, and the model gets updated. Then you test whether the corrected URDF makes the end effector, the tool tip, land closer to the target on a paper task.

Why This Is a Good Topic

This is a strong science fair topic because you can measure real error, change one part of the model, and see whether accuracy improves. It connects to factory robots, surgical robots, and low-cost educational arms, all of which depend on precise positioning. You can also do real engineering work here, like calibration design, computer vision, coordinate transforms, and error analysis, without needing a giant budget.

Research Questions

• How does correcting DH parameters change end-effector position error on a repeatable paper drawing task?
• What is the effect of using more ChArUco board viewpoints on the stability of the estimated robot geometry?
• Does a URDF updated from one fixed phone camera improve accuracy more than the factory calibration model?
• To what extent do joint-angle errors explain the gap between commanded and observed end-effector paths?
• Which calibration target size gives the most consistent pose estimates for a small 5-DOF arm?
• How does the choice of error metric, such as mean position error versus maximum path deviation, change the conclusion about calibration quality?

Basic Materials

• 5-DOF SO-100 or Koch arm clone robot kit.
• Smartphone with a good rear camera and tripod or rigid phone mount.
• Printed ChArUco board on matte paper.
• Plain paper sheets for the drawing task.
• Tape measure or ruler with millimeter markings.
• Computer or laptop for processing images and robot files.
• Notebook for logging joint poses, target points, and error values.

Advanced Materials

• Robot arm with repeatable control interface and joint state readout.
• High-resolution fixed camera or calibrated smartphone camera.
• Rigid ChArUco board mounted on a flat substrate.
• Computer with a robotics stack for URDF editing and forward kinematics checks.
• Image processing software for camera pose estimation.
• Python environment for optimization, plotting, and residual analysis.
• Calipers or a precision ruler for physical dimension checks.
• Optional motion capture or external verification system for validation.

Software & Tools

• Python: Processes camera observations, fits calibration parameters, and compares pose errors.
• OpenCV: Detects ChArUco features and estimates camera pose from images.
• ImageJ: Measures drawn path offsets and compares paper traces to target lines.
• ROS 2: Helps test URDF updates and inspect robot coordinate frames.
• NumPy and SciPy: Support matrix math, optimization, and error fitting.

Experiment Steps

1. Define the geometric error you want to correct, such as link length offsets, joint axis tilt, or base placement error.
2. Choose one repeatable camera setup and lock it down so every image uses the same viewpoint and scale.
3. Build a target set of robot poses that spans the arm's workspace instead of clustering in one small region.
4. Fit your calibration model by comparing observed ChArUco board poses with the arm's predicted forward kinematics.
5. Re-publish a corrected URDF and compare old versus new accuracy on the same drawing task.
6. Analyze which error metric best shows improvement, then check whether the fix generalizes to unseen poses.

Common Pitfalls

• Moving the phone between calibration sessions, which changes the camera frame and breaks pose comparison.
• Testing only poses near the center of the workspace, which hides large errors at the edge of the arm's reach.
• Using a wobbly robot base, which makes geometry errors look like calibration failure.
• Mixing joint backlash with DH error, which blurs the source of the position mistake.
• Judging success by one nice-looking drawing instead of a full error table, which can hide bad overall performance.

What Makes This Competitive

A class-level version shows that calibration improves accuracy. A stronger version proves why it improves and where it still fails. You can push this by testing multiple error models, using unseen validation poses, and reporting uncertainty on every estimate. A really strong project also compares several metrics, like average position error, path smoothness, and worst-case tip deviation.

Project Variations

• Test whether calibration works better on a flat tracing task or on a point-to-point tapping task.
• Compare a fixed phone camera with a fixed overhead webcam as the measurement source.
• Add a second error model, such as camera extrinsic correction, and test whether it improves the URDF update further.

Learn More

• OpenCV Documentation: Search the official OpenCV docs for ArUco and ChArUco pose estimation tutorials.
• ROS 2 Documentation: Read the official ROS 2 robot description and URDF resources for model structure and frame conventions.
• MIT OpenCourseWare, Robotics: Search MIT OpenCourseWare for robotics lectures on forward kinematics, inverse kinematics, and coordinate transforms.
• PubMed: Search for review articles on robot calibration, computer vision-based pose estimation, and manipulator accuracy.
• IEEE Xplore: Search for peer-reviewed papers on manipulator kinematic calibration and vision-based robot error correction.