Currently, it sees a board of 3x3 states, as in: _ _ G G B B _ _ G A next step would be to use real computer vision over a 3D camera in the environment alig

Currently, it sees a board of 3x3 states, as in: _ _ G G B B _ _ G A next step would be to use real computer vision over a 3D camera in the environment aligned to R2D@ and a deep neural network to interpret pixels instead of states. DeepMind has a paper describing this approach: http://arxiv.org/pdf/1312.5602.pdf http://arxiv.org/pdf/1312.5602.pdf