TRON game strategy using deep learning

OData support
Dr. Harmati István
Department of Control Engineering and Information Technology

The computer game called TRON filled the arcades in the 1980s, and its legacy lived on in numerous later editions. The aim of the game is to capture the most possible tiles exactly one time without crossing the opponent's path or the tiles that was occupied before. The strategies used in TRON could be very versatile and complex because you have to keep track of both player's position.

The focus of this thesis is to create an artificial intelligence based control algorithm, which can be applied to TRON. The algorithm uses deep learning based neural networks, today's most hyped machine learning method.

The ideas behind the construction of the neural networks will be focused in the work. It shows each individual steps towards the final structure. As the network can learn from its previous experiments, we can call define different generations of the ultimate algorithm. Due to the better understanding of the learning process the different generations have been implemented individually.

The thesis also contains a TRON simulator, in which the user can be tested against each and every AI algorithm. Thanks to the graphical environment the matches of the different artificial intelligence based strategies can be observed.

The main expectations of the deep learning based strategy was to beat the random algorithm confidently and to be able to learn from its opponents. The results of the games against random algorithm will be detailed, as well as the outcome of the duels between different generations .


Please sign in to download the files of this thesis.