Modular Robot Shape Detection Project

ISEF Category: Robotics and Intelligent Machines

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

The Hook

A robot can lose track of its own shape faster than you might think. If you change the way the cubes connect, the same motor commands can send the end piece somewhere else. That makes self-awareness a real engineering problem, not just a sci-fi idea. Your project asks how a robot can figure out its body, then move with the right math.

What Is It?

This project studies a modular robot made from several cube-shaped units. Each cube has a servo inside, which is a motor that rotates to a set angle. The cubes connect with magnets, so you can reconfigure the robot into different chains, bends, and branches. The hard part is that the robot does not just need to move, it needs to know what shape it currently has before it can predict where its end point will go.

Think of it like a person trying to reach for a cup while blindfolded, but their arm keeps changing length and joint order. The robot uses hall sensors, which detect magnetic fields, to read signature patterns between modules. Those patterns act like clues. The robot then updates its forward-kinematics model, which is the math that predicts where each part should end up after the joints move.

Why This Is a Good Topic

This makes a strong science fair topic because you can test one clear idea, whether sensor signatures can identify robot shape fast and accurately enough to improve motion prediction. You can compare different shapes, sensor layouts, and learning rules, then measure error in the predicted end position. That gives you real data, not just a cool demo. You also learn core robotics ideas like kinematics, calibration, control, and model updating.

Research Questions

• How does the number of connected cubes affect the accuracy of shape identification from hall-sensor signatures?
• How does changing the robot’s geometry change the forward-kinematics prediction error?
• What is the effect of sensor placement on the robot’s ability to distinguish similar shapes?
• To what extent does online model updating reduce end-effector position error after reconfiguration?
• Which classification method best matches a robot shape from hall-sensor data?
• How does magnetic coupling strength affect the stability of shape recognition during motion?

Basic Materials

• Modular robot cubes with servos and hall sensors.
• Magnetic connectors or couplers for each module.
• Microcontroller board such as Arduino, ESP32, or Raspberry Pi Pico.
• USB cable and computer for code upload and data logging.
• Digital caliper for checking module dimensions and joint offsets.
• Tape measure or ruler for end-point position checks.
• Cardboard, grid paper, or a marked floor mat for motion tests.
• Smartphone or camera for recording robot poses.
• Notebook or spreadsheet for tracking shape labels and errors.

Advanced Materials

• Modular robot hardware with encoder feedback and hall-sensor readout.
• Motion capture system or overhead camera setup for pose measurement.
• Calibration jig for repeatable module alignment.
• Bench power supply for stable robot power during repeated trials.
• IMU modules for comparing sensor fusion approaches.
• 3D printer or CNC access for custom module housings and mounts.
• Breadboard, jumper wires, and level shifting components if needed.
• Reference targets or fiducial markers for computer vision tracking.
• Data acquisition interface for synchronized sensor logging.

Software & Tools

• Python: Cleans sensor data, trains classifiers, and calculates kinematics error.
• Jupyter Notebook: Helps you explore patterns in shape labels and prediction results.
• OpenCV: Tracks robot poses from video or overhead images.
• ImageJ: Measures joint angles or positions from frames when simple vision tracking is enough.
• Arduino IDE: Uploads firmware and logs sensor readings from the robot.

Experiment Steps

1. Define the exact shape set your robot will recognize, then make sure each shape is physically repeatable.
2. Choose the sensor features that will represent each module connection, then plan how you will store them as training data.
3. Build a baseline kinematics model for one known shape, then decide how you will measure prediction error after reconfiguration.
4. Design a shape-classification pipeline that maps hall-sensor signatures to robot configuration labels.
5. Plan an online update rule that lets the model adjust when the robot changes shape without starting over.
6. Set controls that separate sensor noise, magnetic variation, and true shape changes so your results stay meaningful.

Common Pitfalls

• Using only a few robot shapes, which makes the classifier look better than it really is.
• Letting magnetic coupling vary from trial to trial, which changes the hall-sensor signature even when the shape stays the same.
• Ignoring servo backlash and joint slack, which can dominate the kinematics error you think comes from the model.
• Training on one module order and testing on another, which causes the system to confuse shape identity with wiring or placement order.
• Measuring end-point position from shaky video or poor camera angle, which hides the real effect of online model updating.

What Makes This Competitive

A stronger project does more than prove the robot can recognize one shape. It compares multiple recognition methods, tests hard-to-tell-apart configurations, and reports confidence, not just accuracy. You can also study how well the model adapts after repeated reconfiguration, since adaptation speed matters in real modular robotics. Careful error analysis, clean controls, and a clear comparison against a baseline model can move the work from a demo to real research.

Project Variations

• Test whether the same shape classifier still works when you swap to cubes of different sizes or magnet strengths.
• Compare hall-sensor shape recognition with camera-based pose estimation to see which method gives lower kinematics error.
• Study whether adding one extra sensor on each module improves online shape discovery more than adding more training data.

Learn More

• NASA Technical Reports Server: Search for modular robotics, self-reconfigurable robots, and kinematics papers from NASA-funded research.
• PubMed: Search review articles on sensor fusion, robotic perception, and adaptive control methods.
• MIT OpenCourseWare: Look for robotics, control systems, and kinematics course notes and assignments.
• IEEE Xplore: Find peer-reviewed papers on modular robots, shape recognition, and online model adaptation through your school or public library access.
• OpenStax University Physics: Use the mechanics chapters for a clear refresher on vectors, rotation, and motion planning.