A hybrid approach to planning and execution in dynamic environments through hierarchical task networks and behavior trees

Abstract

Intelligent autonomous agents that are acting in dynamic environments in real-time are often required to follow long-term strategies while also remaining reactive and being able to act deliberately. In order to create intelligent behaviors for video game characters, there are two common approaches – planners are used for long-term strategical planning, whereas Behavior Trees allow for reactive acting. Although both methodologies have their advantages, when used on their own, they fail to fully achieve both requirements described above. In this work, we propose a hybrid approach combining a Hierarchical Task Network planner for high-level planning while delegating low-level decision making and acting to Behavior Trees. Furthermore, we compare this approach with a pure planner in a multi-agent environment.