Modular behavior trees: language for fast AI in open-world video games

Abstract

As the graphical representation of computer games gradually becomes comparable to cinema, previously neglected non-graphical game aspects start to play an important role in maintaining believability of the game world. One such aspect is the behavior of non-player characters (NPCs), which should appear intelligent and purposeful in the ideal case or not completely stupid at the very least. A good and fast language to express the behaviors is vital for success. We present a visual agent language inspired by behavior trees that was approved for deployment at the core of AI system for an upcoming high-budget open world game.