Marble Maze-Solving Robot is a final-year capstone project designed and built at Monash University by a team of three engineering students over two semesters. The goal: create a robot that could autonomously solve physical marble mazes — and beat humans at their own game.
🎯 Project Overview
Inspired by the classic handheld maze game, this project transformed spatial reasoning and motor coordination into a robotics challenge. The final product combines:
- 🔍 Real-time machine vision using a Raspberry Pi camera
- 🤖 3-DOF Stewart platform with Dynamixel AX-12A servo control
- 📡 Embedded control via PSoC interfacing
- 🧭 A* pathfinding with robust ball tracking and PID movement logic
- 🎮 Manual vs. Autonomous play modes with a touchscreen interface
Users can either manually guide the marble or compete against the robot in a timed challenge. A built-in leaderboard displays the fastest solves — with the robot outperforming human participants in 80% of trials.
📄 Read the full final report:
🖼️ Project Poster:
Click to view full resolution poster.
🛠 Technical Highlights
| Feature | Description |
|---|---|
| Control System | PID-based servo control with 12-direction tilt interpolation |
| Vision | RGB masking for ball and marker tracking with real-time camera feed |
| Platform | 3-leg triangular Stewart platform with interchangeable mazes |
| UI | Touchscreen GUI built in Glade/GTK for selecting modes and viewing results |
| Pathfinding | Optimized A* algorithm using skeletonization for computational speed |
| Hardware | Raspberry Pi 4, PSoC 5LP, Dynamixel AX-12A, MCP3008 ADC, 2-axis joystick |
🧩 How It Works
🧠 Vision + Path Planning
A top-mounted Raspberry Pi Camera detects the marble and maze endpoints using colour-coded stickers. The robot uses an A* pathfinding algorithm to compute the optimal route and adjusts dynamically if it goes off-course.
🤖 Movement & Control
The platform can tilt in 360° using a 3-servo triangular layout. Movements are calculated using interpolated PID control to adjust the marble’s speed and trajectory. The robot handles corners, avoids walls, and corrects errors using techniques like:
- Jitter correction if stuck
- Waypoint lookahead to skip past overshoots
- Speed scaling at turns to prevent collisions
🎮 Human vs Robot
Users can toggle between modes using the touchscreen. In manual mode, a 2-axis joystick provides real-time control. The screen updates with timers and saves performance to a leaderboard — directly comparing the robot’s solution with the human’s.
📈 Results
In real-world trials with 10 participants on Maze 1:
- 🧠 Robot avg. time: 43.77s
- 🧍♂️ Human avg. time: 59.96s
- 🏆 Robot won 80% of the time
The robot’s precise control and vision system led to significantly faster and more reliable performance — even in complex, curved maze layouts.
🌱 Impact & Education
More than a technical build, this robot serves as an interactive STEM education tool, introducing students and the public to:
- Robotics & control theory
- Real-time image processing
- Embedded systems integration
- Engineering project management
The robot has been showcased in demonstrations and poster competitions, sparking interest and engagement among peers and educators.
🔄 Future Work
Planned enhancements include:
- Automatic ball reloading
- Maze holes and fail conditions
- Integration of neural networks for adaptive solving
- Enclosed camera system for improved calibration
🧪 Development Timeline
🗓️ Semester 1
- Research and simulation
- Motion modeling & early prototyping
🗓️ Semester 2
- Full system integration
- Machine vision and GUI development
- Testing and user trials
👥 Team & Acknowledgements
Developers: Travis Barton, Chansophorn Panha In, Mark Jakubowicz
Supervisor: Michael Zenere
Unit: ENG4702 – Final Year Project
Special thanks to the Design & Build Studios and Robogals for ongoing support and educational outreach.
📄 Full Report: