Neural Networks Training for Weapon Selection in First-Person Shooter Games

First person shooters is probably the most well known genre of the whole gaming industry. Bots in those games must think and act fast in order to be competitive and fun to play with. Key part of the action in a first person shooter is the choice of the right weapon according to the situation. In this paper, we propose a weapon selection technique in order to produce competent agents in the first person shooter game Unreal Tournament 2004 utilizing the Pogamut 2 GameBots library. We propose the use of feedforward neural networks trained with back-propagation for weapon selection and we show that there is a significant increase at the performance of a bot. Moreover, we investigate how the performance of the proposed bot is furthermore improved when trained against more difficult enemies.