This project is an attempt to represent and solve the classic Rubik's cube

To solve the problem of a Rubik's cube, we implemented feature-based Q-Learning, a powerful reinforcement learning technique, as well as the utilization of a pattern database, to quantify the quality of near-finished cubes

this project assumes that a Rubik's cube can only execute 180 degree side turns, which greatly reduces the branching factor of the cube's state space tree

this project also attempts to reach a solution for a solved cube which has had n random moves executed on it, this value n currently can be up to 5 to find a goal state for each execution, or in the range of 6-10 to be somewhat successful

puzzle.py includes the state representation of a Rubik's Cube, State(), as well as a few auxillary functions that are used elsewhere in the application

tests.py and others.py includes a variety of test cases that can be executed to confirm the valid implementation of the cube's state represention

Agent.py includes the implementation of a reinforcement learning agent, which executes iterations of Q-Learning, and additionally uses a pattern database, to build up a Q-Table, which is used to eventually try to solve the given random Rubik's cube

clone the repository and verify you have python3 installed

>>> python Agent.py (will run Q-Learning on a n=5 scrambled cube, and attempt to solve it)

to change the number of moves to apply to the initial cube, change n=5 to any value on the line 22 in Agent.py

>>> self.start_state = cube if cube is not None else n_move_state(n=5)

McAleer, Stephen. Solving the Rubik's Cube without Human Knowledge. https://arxiv.org/pdf/1805.07470.pdf

Lapan, Max. Reinforcement Learning to solve Rubik’s cube (and other complex problems!). https://medium.com/datadriveninvestor/reinforcement-learning-to-solve-rubiks-cube-and-other-complex-problems-106424cf26ff