Indexing and retrieving abstract planning knowledge

Abstract

Intelligent systems can reason about plan failures and other critical planning situations by retrieving and instantiating abstract knowledge structures that characterize those situations. This approach requires a vocabulary of thematic abstractions for representing and indexing critical planning situations, and a mechanism for recognizing situations to which the thematic abstractions apply. This dissertation presents a taxonomization of critical planning situations and an abstract indexing vocabulary, both derived from the study of human planning knowledge in the form of common advice-giving proverbs. It also present a parallel, incremental model of retrieval that interleaves and integrates the processes of abstract feature detection and memory search. Together, the vocabulary and the retrieval mechanism constitute a model of memory-based reasoning about the diagnosis and repair of plan failures.