Quantitative Evaluation of Blocksworld Planners

Bibliographic Information

Other Title
  • ブロックスワールドを用いた各種プランナの定量的評価

Search this article

Description

This paper evaluates the available planners such as STRIPS, NONLIN, TWEAK by using so called the blocksworld problem. Many planners have been proposed so far. However, the research on these planners only describes the mechanisms of making plans, and their performances are not well known. In this paper, we first define the blocksworld problem. Using this problem, we evaluate the three planners in terms of plan length and CPU time. Experiments will show the following. (1)The plan length and CPU time of the linear planner, STRIPS, get worse exponentially as the interaction between subgoals increases. (2)The nonlinear planner, NONLIN, has better performance than STRIPS in terms of the plan length. However, the CPU time of NONLIN also gets worse exponentially for a large problem. (3)The nonlinear and least commitment planner, TWEAK, has more interaction-removal methods than NONLIN. However, TWEAK can solve only small problems for lack of the knowledge about how to use the methods. We will also show that blocksworld problems have been used to illustrate many planners, however, they are not appropriate for evaluation because there are little differences between trivial solutions and optimal solutions.

Journal

Details 詳細情報について

  • CRID
    1570009752154508544
  • NII Article ID
    110002941074
  • NII Book ID
    AN10464118
  • ISSN
    09196072
  • Text Lang
    ja
  • Data Source
    • CiNii Articles

Report a problem

Back to top