Comparison of Different Artificial Intelligence Techniques Applied in a Multiplayer Shooter

Abstract

Lately, video games sector is experiencing an exponential growth based on the appearance of a great quantity of new multiplayer video games. In fact, this feature has turn basically into a requirement. Most video games provide a story mode in which all narrative action is performed and in which the player learns the mechanics and learn how to master the controls and a multiplayer mode, in which players carry out the playful or ludic action based on the competition and the showing of their skills. This duality may be caused by the fact that in multiplayer mode, players match up each other without the intervention of any agent controlled by an artificial intelligence. This supposes a greater challenge due to the fact that, usually, is very common to determine the patterns with which an enemy has been built after a few attempts to beat it. Once a player has discovered its behavior, the complexity of the battle is drastically reduced, and so, the fun degree. This paper constitutes the memory of the Final Project in the Game Design and Development degree and proposes a solution for this existent problem, by means of the design and implementation of machine learning techniques, specifically, reinforcement learning. With this solution, NPCs (non-playable characters) are able to learn from the player’s actions and modify its behavior to provide a better experience to the gameplay. In this project, two different types of enemies have been developed with Unreal Engine 4 for a shooter video game called Hive: Altenum Wars, which is expected to be released in a few months. On the one hand, there are the agents built up with predefined rule-based artificial intelligence techniques, specifically, behavior trees. On the other hand, analogous agents have been developed based on reinforcement learning to provide them the ability to adapt their behavior to the player’s gaming experience.