Evolutionary neural network for ghost in Ms. Pac-Man

Ms. Pac-Man is a popular chasing and evading game and the ghost character in the game is controlled by script. This article evolved an evolutionary neural network for the red ghost to chase Pac-Man. Red ghost´ position, Pac-Man´s position and Pac-Man´s state are considered to be the inputs of the neural network, and the output is the direction of Red ghost to move in the next step. We also proposed a fitness function to raise capture ability in evolution so that the Red ghost learns by itself in simulation. Experimental results show that the agent learns well and plays better in teamwork than the traditional script controlled ghost.