Abstract:

Video games are simplified representations of complex real-world environments and consist of several simplified tasks of real-world problems. Video games are also good testbeds for artificial intelligence algorithms. Solving the tasks in video games might lead to solve the real-world problems using tested artificial intelligence algorithms. In this work we developed an artificial game playing agent using Neuro Evolution of Augmenting Topologies for Flappy Bird game. Game states are represented using eight features including distance of the bird from ground and to pipes, velocity of the bird and top and bottom locations of the pipes. Experiments show that our agents are able to learn the game environment without prior knowledge of the game domain.

Keywords: game artificial intelligence, neuroevolution, flappy bird, video games