Ready Player AI

The rapid development and increasing use of artificial intelligence in recent years has shown how powerful and useful this technology is.

This project’s goal was to get a better understanding of this technology by self-developing  machine learning algorithms and implementing these into two  multiplayer games: “Pong”, a simple, tennis like, game, and "Footsies", a 2D fighting game. Next to typical  software coding challenges connected to the development of AI algorithms, a key problem in implementing AI was that games do not provide pre-classified data. The AI relies instead on a loop of exploration and feedback on the results the chosen actions led to. This is the basic idea behind the two different algorithms I applied, a Genetic Algorithm and the Actor-Critic Method, to train the Artificial Neural Networks developed for this project.

For Pong, both the Genetic Algorithm and Actor-Critic Method achieved near perfect results and the AI player performed substantially  above human standards. The implementation for Footsies only achieved moderate success. In Footsies, both algorithms struggled with using more complex strategies and got stuck after having achieved initial improvements. Further research and optimization would be required to achieve human-level performance.