EffBT: An Efficient Behavior Tree Reactive Synthesis and Execution Framework

Behavior Trees (BTs), originated from the control of Non-Player-Characters (NPCs), have been widely embraced in robotics and software engineering communities due to their modularity, reactivity, and other beneficial characteristics. It is highly desirable to synthesize BTs automatically. The consequent challenges are to ensure the generated BTs semantically correct, well-structured, and efficiently executable. To address these challenges, in this paper, we present a novel reactive synthesis method for BTs, namely EffBT, to generate correct and efficient controllers from formal specifications in GR(1) automatically. The idea is to construct BTs soundly from the intermediate strategies derived during the algorithm of GR(1) realizability check. Additionally, we introduce pruning strategies and use of Parallel nodes to improve BT execution, while none of the priors explored before. We prove the soundness of the EffBT method, and the experimental results demonstrate its effectiveness in various scenarios and datasets.