Is "early commitment" in plan generation ever a good idea?

  • Authors:

  • David Joslin;Martha E. Pollack

  • Affiliations:

  • Computational Intelligence Research Laboratory, University of Oregon, Eugene, OR;Department of Computer Science and Intelligent Systems Program, University of Pittsburgh, Pittsburgh, PA

  • Venue:
  • AAAI'96 Proceedings of the thirteenth national conference on Artificial intelligence - Volume 2
  • Year:
  • 1996

Quantified Score

Hi-index 0.00

Visualization

Abstract

Partial-Order Causal Link planners typically take a "least-commitment" approach to some decisions (notably, step ordering), postponing those decisions until constraints force them to be made. However, these planners rely to some degree on early commitments in making other types of decisions, including threat resolution and operator choice. We show why existing planners cannot support full least-commitment decisionmaking, and present an alternative approach that can. The approach has been implemented in the Descartes system, which we describe. We also provide experimental results that demonstrate that a least-commitment approach to planning can be profitably extended beyond what is done in POCL and similar planners, but that taking a least-commitment approach to every planning decision can be inefficient: early commitment in plan generation is sometimes a good idea.