Comparing the Real-Time Performance of a PID-Controller and Reinforced Learning-Based Labyrinth Board Game : Construction and Optimization of a Line-Following Marble on aSelf-Balancing Platform

Sammanfattning

This thesis explores the development of an autonomous system capable of solving the Brio Labyrinth game using computer vision and control theory. The project involved integrating a camera system with image processing algorithms to continuously track the position of a rolling marble. The plane on which the marble moves is actuated by two stepper motors, enabling precise tilting in both axes. A Python-based computer vision module communicates with an Arduino microcontroller to implement real-time closed-loop control. This division of tasks allowed for efficient coordination between software and mechanical subsystems. The resulting system demonstrates the feasibility of using a vision-based feedback loop to control a physical, two-axis labyrinth platform. The project illustrates key concepts in automation, mechatronics, and embedded control, and serves as a foundation for future exploration of adaptive or learning-based control strategies