From: Christophe Bisciglia (chrisrb_at_cs.washington.edu)
Date: Tue Apr 15 2003 - 11:06:36 PDT
S. Kambhampati, Refinement planning as a unifying framework for plan
synthesis
This paper summarizes all of the current major strategies for planning and
shows how, at an abstract level, each of them is a form of refinement.
The most important ideas form this paper are quite abstract. First off,
the paper presents an allegedly .brand-free. way of describing planning
strategies. It then shows how each of these strategies is a form of
refinement. Essentially, whether the planner uses a plan space or a state
space, an iterative process is applied that brings the set of plans being
considered closer to the set of solutions.
Another idea I liked from this paper was naming parameters common (again
at an abstract level) to all planning strategies. This allowed for a
pseudo-objective comparison of the advantages and disadvantages of
different strategies on some common ground.
Nevertheless, the flaws in this paper come directly from it being so
abstract. The paper is so general, and uses so much new nomenclature so it
is hard to see how many of the ideas manifest themselves in reality. It
talks about planning in such a general sense that I am left saying, .well
duh?. . Kevin phrased this perfectly by essentially saying .while I have a
plan set that isn.t the solution, refine the plan set..
Despite the abstract level of the paper, it still manages to propose
several research directions. At the end of the paper, the author points
out that there are only two fundamental refinement strategies (plan space
and state space). It is suggested that there may be a new, novel approach
of refinement that works better than both.
This archive was generated by hypermail 2.1.6 : Tue Apr 15 2003 - 11:06:37 PDT