Self-Repairing Rover With Adaptive Wheel Control

ISEF Category: Robotics and Intelligent Machines

Subcategory: Other  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

A robot can lose a wheel and still keep going. That sounds like sci-fi, but the key clues are already on the robot, in motor current and motion data. If you can spot failure fast, you can teach the rover to compensate before it stalls. That makes this a strong project for learning fault detection and online control.

What Is It?

This project studies fault-tolerant robotics. That means the robot does not just drive, it notices when something goes wrong and changes its own behavior. In your case, a current sensor like the INA219 can detect a wheel that starts drawing unusual power, and an IMU, which measures motion and tilt, can detect left-right imbalance.

Think of it like walking with a twisted ankle. You do not keep moving the same way. You shift your weight and change your stride. A rover can do something similar by adjusting motor commands, steering bias, or wheel speed ratios. Bayesian optimization is one way to search for better control settings using feedback from each trial. It tries a set of settings, checks the result, and then picks smarter settings next time.

Why This Is a Good Topic

This is a good science fair topic because you can measure a clear before-and-after effect, recovered speed, stability, or both. You also have a real engineering problem to solve, since robots in the field can lose traction, damage a motor, or run on uneven ground. You can learn fault detection, sensor fusion, control tuning, and optimization, all in one project. Those are the same skills used in serious robotics work.

Research Questions

• How does wheel failure change rover velocity, current draw, and heading error compared with normal driving?
• What is the effect of using current spike detection alone versus current plus IMU asymmetry for fault detection accuracy?
• Does Bayesian optimization recover rover speed better than fixed manual control adjustments after one wheel fails?
• To what extent does the terrain type change how well the rover can compensate for a broken wheel?
• Which control parameters most improve recovered velocity without raising energy use too much?
• How does the time to detect failure affect the final recovered speed?

Basic Materials

• Small four-wheel rover chassis with DC motors or gear motors.
• Motor driver board compatible with your rover.
• INA219 current sensor or similar current sensor.
• IMU sensor such as an accelerometer and gyroscope module.
• Microcontroller such as Arduino, ESP32, or Raspberry Pi Pico.
• Battery pack with safe power regulation.
• Breadboard and jumper wires.
• Test surface with at least two terrain types, such as smooth floor and carpet.
• Digital scale or simple load materials for controlled wheel damage tests.
• Laptop for coding and data logging.

Advanced Materials

• Research-grade rover chassis with modular wheel mounting.
• High-resolution current sensing hardware.
• 6-axis or 9-axis IMU with stable sampling.
• Motor encoder system for wheel speed measurement.
• Data acquisition interface for synchronized logging.
• Adjustable load rig for controlled wheel failure simulation.
• Motion capture access or overhead tracking camera system.
• Safety-rated battery management hardware.
• Bench power supply for debugging.
• Mechanical tools for repeatable wheel damage or wheel lock simulation.

Software & Tools

• Python: Runs the control loop, logs sensor data, and fits the Bayesian optimizer.
• Jupyter Notebook: Helps you clean data, plot recovery curves, and compare trials.
• NumPy: Stores sensor arrays and handles numerical calculations.
• SciPy: Supports statistics, filtering, and optimization comparisons.
• OpenCV: Tracks rover motion from video if you want an extra ground-truth measurement.

Experiment Steps

1. Define the failure you will study, such as a stalled wheel, a slipping wheel, or a locked wheel, so your test stays repeatable.
2. Choose the performance metric you will optimize, such as recovered velocity, heading stability, or energy use, and decide how you will measure it.
3. Design a baseline controller first, so you can compare your adaptive method against a fixed behavior.
4. Plan the detection logic that flags failure from current and IMU signals, then decide what counts as a true alarm.
5. Build the optimization loop that proposes new gait settings after each trial and records whether the rover improved.
6. Set up a fair comparison across terrain types and summarize which control strategy restored motion fastest.

Common Pitfalls

• Treating every current spike as wheel failure, which creates false alarms when the rover turns or starts moving.
• Changing battery level between trials, which shifts motor current and hides the real signal.
• Using the IMU without calibrating sensor bias, which makes asymmetry look larger or smaller than it really is.
• Letting the rover drift onto different floor surfaces between runs, which adds noise to the velocity comparison.
• Testing too many control settings at once, which makes it hard to tell which change actually improved recovery.

What Makes This Competitive

A stronger version of this project would separate detection, adaptation, and recovery into clear measurable stages. You can raise the level by proving that your method works across more than one failure type, not just one broken wheel. You can also compare Bayesian optimization with simpler search methods or fixed heuristics to show why your approach learns faster. Careful statistics, repeated trials, and clean ground-truth motion tracking will make the results much stronger.

Project Variations

• Test the same recovery system on a two-wheel differential rover instead of a four-wheel platform.
• Replace the wheel fault with deliberate traction loss on one side, then compare how the controller reacts.
• Add overhead video tracking and compare vision-based recovery metrics with IMU-based recovery metrics.

Learn More

• NIH PubMed: Search for review articles on fault-tolerant robot control, adaptive locomotion, and Bayesian optimization in robotics.
• NASA NTRS: Search the NASA Technical Reports Server for rover locomotion, autonomous fault detection, and terrain traversal studies.
• MIT OpenCourseWare: Find robotics and control courses for background on feedback control, state estimation, and system identification.
• IEEE Xplore: Read abstracts and open-access papers on fault detection and adaptive robot gait control.
• arXiv: Search for recent preprints on Bayesian optimization for robotics and online adaptation.