Managing the life cycle of plans

Abstract

The scalability of recent planning algorithms allows developers to automate planning tasks, which so far have been reserved to humans. However in real-world applications, synthesizing a plan is just the beginning of a complex life-cycle management process. Plans must be organized in large collections, where they can be grouped along different purposes and are amenable to the search, inspection, evaluation, and modification by human experts or automated reasoning systems. Eventually, plans will outlast their utility and be replaced. We present our solution to plan life cycle management for an autonomic computing application. We focus in particular on the automatic synthesis of plan metadata for plans containing conditional and parallel actions, well-structured loops, and non-deterministic choices. The plans are of unknown origin, i.e., their underlying action model, which could provide us with pre- and postconditions, is not available. New analysis techniques are presented that uniformly generate metadata for plans, thus allowing a system to embed plans into context and organize them in meaningfully structured plan repositories. Copyright © 2005, American Association for Artificial Intelligence (www.aaai.org). All rights reserved.