Building behavior trees from observations in real-time strategy games

This paper presents a novel use of motif-finding techniques from computational biology to find recurring action sequences across many observations of expert humans carrying out a complex task. Information about recurring action sequences is used to produce a behavior tree without any additional domain information besides a simple similarity metric - no action models or reward functions are provided. This technique is applied to produce a behavior tree for strategic-level actions in the real-time strategy game StarCraft. The behavior tree was able to represent and summarise a large amount of information from the expert behavior examples much more compactly. The method could still be improved by discovering reactive actions present in the expert behavior and encoding these in the behavior tree.