From: MAUSAM (mausam_at_cs.washington.edu)
Date: Tue May 20 2003 - 11:21:05 PDT
A Knowledge-based approach to planning with incomplete information and
sensing : Petrick, Bacchus
This paper presents a representation of a belief with disjunctive
uncertainty as a knowledge base. Actions now map from a knowledge base to
another knowledge base. And so the solver has the search in this knowledge
base space.
The various types of predicates it encodes are conjunctive clauses. The
atomic clauses included are whether a formula is true, whether the agent
knows that a formula is exactly true in all worlds or false in all worlds,
whether the agent knows a function value and xor on other formulas. Note
that this representation is a restricted version of all the knowledge
rules that could be represented. For example it does not handle
disjunctions. But this representation includes functions and runtime
variables.
The planner grows a contingent plan by adding new actions or new
branches in this knowledge base. They don't discuss much on how the search
is to be done. I think some extensions are possible here.
Moreover, the choices they make about the kind of representations they
allow in the database puts a bound on the type of problems that can be
solved using this method. They don't give any clear rationale why they
choose one representation over others possible. One could study this
extensions and compare the new problems that become solvable and how much
more time is taken to solve the previous ones.
Moreover, usefulness of this method can only be demonstrated once it is
compared with the other existing planners.
Also, this approach only works for disjunctive uncertainty. I wonder how
one might apply similar approach with probabilistic uncertainty, something
like probabilistic databases.
This archive was generated by hypermail 2.1.6 : Tue May 20 2003 - 11:21:10 PDT