Knowledge-based Execution Configuration for Adaptive Behavior Trees

Abstract Automated planning is commonly used to obtain plans to solve particular tasks. To execute these plans, Behavior Trees have emerged as a popular execution architecture due to their reactivity and modularity. Configuring the execution of a plan into a Behavior Tree requires expanding the high level actions into the proper Behavior Tree structure. Typically, this is achieved using of pre-defined rigid templates. In this work, we propose a novel approach to generating the Behavior Trees using ontologies. The generated Behavior Trees are tailored to the specific requirements of a task and the world by using modifiers to a base template that provides a general solution to the task. These modifiers and their properties are formally defined using ontologies. A proof of concept is developed, illustrating how the Behavior Trees for the execution of manipulation tasks in a kitchen scenario can be adapted to varying conditions by applying different modifiers.