Hmm could machine learning be used to make a general twisty puzzle solver?
Use machine learning to estimate a distance from solved for any state of the puzzle. Then test all random twists from the current state to find which one brings the puzzle closest to solved. Stuff like this has kinda been done but it'd be interesting if it could be trained with multiple puzzles and still work.
Use machine learning to estimate a distance from solved for any state of the puzzle. Then test all random twists from the current state to find which one brings the puzzle closest to solved. Stuff like this has kinda been done but it'd be interesting if it could be trained with multiple puzzles and still work.
