Using Abstraction and Nondeterminism to Plan Reaction Loops

Abstract

By looping over a set of behaviors, reactive systems use repetition and feedback to deal with errors and en-vironmental uncertainty. Their robust, fault-tolerant performance makes reactive systems desirable for exe-cuting plans. However, most planning systems cannot reason about the loops that characterize reactive sys-tems. In this paper, we show how the structured ap-plication of abstraction and nondeterminism can map complex planning problems requiring loop plans into a simpler representation amenable to standard plan-ning technologies. In the process, we illustrate key recipes for automatically building predictable reactive systems that are guaranteed to achieve their goals.