Theory and Practice in Planning

This paper explores opportunities for synergy between theoretical and applied researchers in the area of automated planning. From the diiering perspectives of theorists, experimentalists, and practitioners, we examine the causes of current dissatisfaction with existing theoretical and experimental results. We survey approaches from theoretical computer science that provide alternatives to the standard measures of performance, speciically asymptotic worst-case analyses. As an alternative to worst-case analyses relying on all-powerful, all-knowing adversaries, we investigate average-case performance measures and the properties of distributions governing the generation of problem instances. Regarding asymptotic arguments, we consider sharp-threshold functions, the relevance of phase-transition phenomena, and the statistical properties of graphs of small order. The resulting perspective emphasizes particular problem instances and speciic algorithms rather than problem classes and complexity results that pertain to all algorithms of a given order of growth. We issue a challenge to theorists, experimentalists, and practitioners alike to raise the level of expectation for collaborative scientiic research in planning and suggest some initial research directions. We identify existing results in planning and automated reasoning that we believe point the way to further progress. Finally, we present some preliminary results that demonstrate the potential for synergy involving theoretical and empirical studies.