Two Human-Like Imitation-Learning Bots with Probabilistic Behaviors
Chris Pelling, Henry Gardner · 2019 IEEE Conference on Games (CoG) · 2019
We present details of two imitation-learning, game-playing agents - "SVMBot" and "PETBot" - that feature probabilistic modelling of some low-level combat behaviours. Both bots used a support-vector-machine approach for aiming and a novel, probabilistic model for jumping behaviours. PET-Bot also used a probability-estimation-tree (PET) technique for combat movement. The bots were developed for the FPS game Unreal Tournament 2004 and one of each was submitted to the qualification round (SVMBot) and the final round (PETBot) of the 2009 2K BotPrize competition where they obtained good results. They were then compared with each other as independent variables in a human-computer-interaction (HCI) between-subjects experiment.