CREATING AN ARTIFICIAL GAME PLAYER FOR MODIFIED STICK RUNNING USING NEUROEVOLUTION OF AUGMENTING TOPOLOGIES (NEAT)

Aldrich S. Goco and Rozano S. Maniaol

Abstract:

In this work, an artificial game player for the Modified Stick Running was created using the NeuroEvolution of Augmenting Topologies (NEAT) algorithm. NEAT was used to evolve the neural network of the game agent to play the game indefinitely. Game states are represented using three features including the height and width of the nearest obstacle and also its distance to the game agent. The fitness of the agent is the combination of its score and traveled distance. The Algorithm was run for 30 trials with a maximum allowable generation of 30 and a maximum scoring of 50 points. The experiment has shown that the NEAT algorithm has a success rate of 83.33 percent and an average of breakthrough at 16.8 generations.