Firstly, thanks for sharing your implementation of DQN on a Rubik's cube - I find it very interesting. I noticed that your cube model (as described in the cube environment and in notebook part 1) only considers the position of each of the cubies. It does not take into account the rotation of the cubies (ie. the corner pieces each have three possible orientations, and the edges have 2 possible orientations).
This means that you are only positioning each of the cubies into the correct location, but not actually solving the cube. I was wondering if you have considered this, as it would mean your algorithm is only solving a greatly reduced version of a Rubik's cube.
Hi Matt,
Firstly, thanks for sharing your implementation of DQN on a Rubik's cube - I find it very interesting. I noticed that your cube model (as described in the cube environment and in notebook part 1) only considers the position of each of the cubies. It does not take into account the rotation of the cubies (ie. the corner pieces each have three possible orientations, and the edges have 2 possible orientations).
This means that you are only positioning each of the cubies into the correct location, but not actually solving the cube. I was wondering if you have considered this, as it would mean your algorithm is only solving a greatly reduced version of a Rubik's cube.