This is an experiment using reinforcement learning and Markov Decision Process to create a
pacman AI. I made a c++ clone of pacman, and implemented this AI on top of it. The synthetic
player has a partial view of the game. It knows five parameters. The first one had four values,
giving the direction of the nearest pellet. The four other dimensions describe the four Pacman
directions. For each direction, Pacman knows whether he is facing a wall, a ghost one or two
step away from him, a free-of-ghosts path less than 18 steps long, or free-of-ghosts path
longer than 18 steps long. You\'ll find a video of this experiment in that section.