Learning to win in a first-person shooter game

Designing a character's behavior in a first-person shooter (FPS) game typically requires a considerable amount of effort due to a complex in-game environment. This paper proposes an efficient strategy of training a character in the Quake III Arena, a FPS game, to be more adaptive and also to have human-like learning capabilities and behaviors. Specifically, a particle swarm optimization (PSO) algorithm is introduced to provide a character with self-learning abilities. Like many computer games, the Quake III Arena utilizes a rule-based system which is constrained by several weights in the software. The effectiveness of these constrained weights is dependent upon the programmer's knowledge about the game. In order to increase a character's learning ability, an efficient method using the PSO algorithm is developed to determine the constrained weights for the behavior control. In the conducted experiments, the PSO algorithm is implemented to design a non-player character which is shown to be superior to the other characters originally created in the Quake III game. This efficient training strategy decreases the amount of effort required by game designers to design an intelligent character's behavior.