From: Stanley Kok (koks_at_cs.washington.edu)
Date: Thu Apr 10 2003 - 21:08:08 PDT
Paper Title: Planning as Heuristic Search: New Results
Authors: Blai Bonet and Hector Geffner
One-line summary:
The paper presents an improved Heuristic Search Planner that performs
regressive search.
Most Important Ideas in the Paper:
1. Regressive search improves the performance of HSP. This reinforces
the efficacy of the commonly used regressive search strategy.
2. Graphplan is a form of HSP. This provides another angle from which
to analyze and improve Graphplan.
Flaw in the Paper:
1. Even though it argues that Graphplan is a form of heuristic
search, the paper does not perform any experiment to bear this out,
nor did it provide a rigorous proof. This gives an incomplete feel
to the paper.
Important, open research questions:
1. HSPr performs well in some domains but not others (e.g. 'grid').
What kinds of domain lend themselves to being solved by HSPr? And
what features makes a domain unsolvable by HSPr? The answers can
guide our application of different planning paradigms to new
problems.
2. Could other planners (in addition to Graphplan) be reduced to
a heuristic search planner? If this is the case, we can focus our
efforts on improving the performance of heuristic search planners.
This archive was generated by hypermail 2.1.6 : Thu Apr 10 2003 - 21:08:09 PDT