Deriving and Using Abstract Representation in Behavior-Based Systems

Abstract

We present a representation that addresses two current limitations of the behavior-based systems (BBS) (Matarić 1992), (Arkin 1998): the lack of abstract representation within behaviors (which makes them hard to use in complex, sequential problems) and the need for behavior redesign even for tasks that use subsets of the same behavior set. We introduce the concept of behavior networks, based on the abstract behaviors representation described below. We distinguish the following two types of behavior preconditions: world preconditions (activate the behaviors based on the state of the environment) and sequential preconditions (task-dependent conditions, often postconditions of other existing behaviors). In standard BBS behaviors, both types of preconditions are tested together, thus hard-coding a particular solution. The key step in adapting specialized behaviors