From: MAUSAM (mausam_at_cs.washington.edu)
Date: Tue May 06 2003 - 09:59:38 PDT
Incremental contingency planning
This paper tries to solve a fairly general planning problem involving
continuous resources, durative actions, continuous outcomes, plan
utilities etc.
As we know, a general planning problem like this is extremely hard to
solve. Kudos to the authors to have actually taken up such a problem. In
the paper, they present an algorithm called Incremental Contingency
Planning, in which they start with a seed plan, find contingency branches
in that plan but not everywhere. They find places, whereby adding
branches would yield a higher expected utility function. They add branches
there. So on the whole, the algorithm runs in three steps : start with a
seed plan, evaluate the value of plan and estimate the value of branch,
and add branches where estimated values of branches are high.
They use planning graph scenario, however, one would have expected that
there will be probabilities somewhere, much like Pgraphplan. It is unclear
if they are handling the deterministic case, or there are probabilities
implicit in the discussion, or did I miss something.
The motivation of the problem is good, however, it still eludes me as to
how this solution will cater to all the difficulties they have mentioned,
like continuous outcomes, durative actions etc. I don't understand where
this solution handles continuous outcomes (for example).
Some parts of the paper are tersely written, which make it a little harder
to understand. I got quite confused in the discussion of utility
propagation. It would have been nicer had they given some theorems about
soundness of their method.
This seems like initial attempts towards solving a hard problem, so it
will be unfair to criticize the authors about results, but they actually
don't show any experiments. It had been nicer, had they at least done some
experimental work to just keep verifying that they are on the right track.
This archive was generated by hypermail 2.1.6 : Tue May 06 2003 - 09:59:39 PDT