Chapter 1. INTRODUCTION
Video games are constantly evolving. Up until recently there has been a surge of research in the field of graphics in the video game industry with the overall goal being real- time photo realistic graphics. However graphics have now become realistic enough that companies are looking at other ways in evolving games and bringing them to the next level. Apart from the visuals, the other main feature of a game that grabs a player’s attention is the gameplay. Gameplay is made up of several different components with one of the most important being artificial intelligence (AI). AI in a game controls the decision making processes of all non-player characters (NPCs) and provides players with challenges they must overcome to progress in the game. With the so-called ‘graphics race’ effectively finished, the field of AI has become a hot-bed of research with new games using more and more advanced AI than ever before. Therefore this dissertation was motivated by the desire to investigate an exciting new AI technology and compare it with what already exists to determine if this new technique really is an improvement.
Finite state machines (FSMs) have long been a faithful servant to the AI programmer due to their ease to program and robustness. NPC’s controlled by FSMs have a number of states and can only be in one of these states at any time in a game. For example, a simplistic FSM could have three states: Roam, Attack and Hide. If the agent is in the Roam state and it sees an enemy it could change to the Attack state. If upon executing the Attack state the NPC discovers that it is low on health, it could change state to Hide and go and find a hiding spot. The AI programmer must decide what states an NPC is going to have and what conditions cause the NPC to move from one state to another. However there are difficulties associated with FSMs that have brought about a desire for cleaner and more flexible technologies.