From: Christophe Bisciglia (chrisrb_at_cs.washington.edu)
Date: Tue May 06 2003 - 11:01:15 PDT
Dearden et al. Incremental Contingency Planning
This paper proposes a method called Just In Case (JIC) planning to form
contingent plans over a continuous space.
The first big idea from this paper was augmenting the .mainline. or .seed.
plan with contingency plans at the most likely points of failure. The
points were determined by considering if an alternate plan at that point
in the mainline could still produce a favorable utility.
The other main point was the notion of utility tables, and propagating
them backwards through the plan graph. A utility table is essentially a
piecewise constant function dependant upon the continuous variables that
define the space. They focused on energy as an example throughout the
paper. These tables can then be merged at points where the plan graph
branches, so from that state, it can be inferred that depending on your
variables, you can hope to achieve certain utility levels.
My gripe with the paper was their extensive use of plan graphs without
clearly justifying their disregard for mutex relationships. It wasn.t
clear why mutex relationships weren.t computed (accept for a brief
allusion that computing resources were very limited on a Mars rover).
Furthermore, although the ideas seem quite plausible, some experiments
would have been helpful, but this does seem to be an idea in its infancy,
so I suppose this is acceptable.
My first gripe leads to an interesting extension . how much better could
mutex relationships make JIC planning? It seems that anything that reduces
the possible plan space would be useful, and it would be interesting to
see if it does so enough to justify their computation. This is something I
would expect to see if the research were more mature.
This archive was generated by hypermail 2.1.6 : Tue May 06 2003 - 11:01:16 PDT