Application of Reinforcement Learning to Solve Rubrik’s Cube with Markov Decision Process

Application of Reinforcement Learning to Solve Rubrik’s Cube with Markov Decision Process

Reinforcement Learning, Rubik’s Cube; Markov Decision Process

The Rubik's Cube is a tricky puzzle that can be arranged in countless ways, making it hard for both people and computers to figure out. While standard solving methods use fixed strategies, this research looks into using reinforcement learning (RL) to create a flexible and effective way to solve it. The goal of this research is to develop an RL-based solver that uses the Markov Decision Process (MDP) system, focusing on speed, efficient moves, and the number of steps needed to solve the cube. The suggested model uses Q-learning and Monte Carlo Tree Search (MCTS) to figure out the best moves at each stage of the game, training through lots of Rubik's Cube simulations. What makes this research unique is the combination of MCTS with Q-learning, which improves decision-making by needing fewer moves than standard methods. The tests show that this model reaches almost perfect solutions with fewer moves, doing better than simple rule-based methods. Also, a web app was created to give live solving techniques based on the cube arrangements that users provide. This research helps grow the use of RL in puzzles like the Rubik's Cube and gives a useful tool for fans who want to get better at solving the cube

Defni1, Andi Fathul Mukminin2, Ainil Mardhiah3, Titin Ritmi4, Junaldi5, Yuhefizar6, Fibriyanti7

https://jurnal.iaii.or.id/index.php/RESTI/article/view/6552/1135

Information Technology, Padang State Polytechnic, West Sumatera, Indonesi

October 8, 2025

FAJAR BAGUS W

PDF

ENGLISH

TEXT