Improving the efficiency of planning

Symbolic planning involves extensive computation for detecting and handling goal interactions. Past planning systems can be classified as either domain-dependent or domain-independent, depending on how they deal with the goal interactions. Domain-dependent systems rely on heuristics that are specific to their particular applications domains, and are of limited generality. Domain-independent systems make use of general planning knowledge independent of any particular domains, but have been mainly inefficient. This thesis presents two approaches to improving the efficiency and broadening the applicability of planning knowledge. Whenever these conditions are satisfied, efficiency can be provably improved. In addition, these conditions enable one to preprocess the planning knowledge of a system before problem solving starts, and offer guidelines for the design of knowledge representation. Second, a set of restrictions are imposed on the goal interactions so that, whenever they are satisfied, efficient planning methods can be developed. The restrictions does not depend on any specific domain knowledge. Therefore the approach is more general than domain-dependent planning. In addition, they offer more knowledge about goal interactions and thus enable more efficient methods to be developed than purely domain-independent approaches.