RubikNet

Deep Reinforcement Learning for Rubik’s Cube Solving with Stepper Motor Apparatus


Project Domains Mentors Project Difficulty
Reinforcement Learning, Python Programming, Robotics Simulation Akash Kawle, Ansh Semwal Hard

Project Description

This project will introduce students to physical systems based RL-policy networks and logic solvers by building a system that can solve a Rubik’s Cube. Starting with a 2x2 cube and progressing to the standard 3x3, students will develop a Deep Reinforcement Learning (DeepRL) agent capable of learning optimal sequences of moves to solve the cube from any scrambled state.
The learned move sequences will be executed by a low-cost robotic setup driven by stepper motors and controlled via an ESP32 microcontroller. Along the way, students will gain hands-on experience in Python, RL frameworks, reward engineering, computer vision for cube state recognition, and embedded systems programming.
By the end of the project, students will have a working pipeline that connects decision-making AI with real-world actuation—a complete software-to-hardware system.


Resources

Inspiration
Intro to RL
Monte-Carlo Tree Search