AI has frequently been criticized for being `stuck in the microworld' because of the common inability of AI systems to cope with the complexity of real domains. Often, adding details removes regularity, transforming a representation from a few simple structures to a large, unwieldy collection of specialized ones.
This paper addresses this problem in the context of representing planning operators (domain-specific knowledge about the effects of actions in a domain) for use by AI planning systems. We present a novel approach in which domain-specific operators are represented as a composition of general components, and show that the problem of manually building a detailed set of operators can be avoided by constructing them from a small number of such components instead. Each component encapsulates information about a domain feature that might be modeled, and each may contribute to several operators. Moreover, we describe how the choice of what to model and what to ignore in a domain can then be easily varied, simply by controlling which components are used. Finally, we show how operator sets built in this way can be used by planning algorithms.
Technical Report AI06-331, University of Texas at Austin.