NEAT implementation for adapting neural networks applied to ATARI Asteroids

Abstract

The NEAT algorithm grants useful tools for creating agents that beat simple games, as the agents created through it can give simple and clear outputs, given a set of defined inputs, as well as a fitness/reward formula that is straightforward to design for simple games. Not only that but it's capabilities of generating variable agent "genomes" which take different approaches to the possible hurdle the game presents. This project focuses on developing agents that are able to play the Atari game, Asteroids, with some level of competence, and how these different species fare against the game's hurdles, including the various elements of randomness that the game's setting presents.