From: Sumit Sanghai (sanghai_at_cs.washington.edu)
Date: Tue May 13 2003 - 11:03:02 PDT
Agent-Centered Search
--Sven Koenig
Summary : The paper talks about the various properties and applications of
agent-centered search
Ideas : The paper talks about the use of agent-centered search in real
time domains such as game playing, robot localization etc where the goal
is not just to minimize the plan execution cost, instead it could be to
minimize the sum of planning + execution cost or minimize execution cost
given that planning can only cnsume a given maximum amount of resource
(time). In general, the idea is to consider algorithms where the planning
and execution are interleaved. In this regard, they talk about LRTA* which
is a variant of A* but works in real time (and local space). They also
talk about other algorithms such as Node Counting, uninformed LRTA* and
Min-Max LRTA* which can be used where there is uncertainty about the
domain.
Flaws : All the algorithms presented seem to be a variant of tehr A*
algorithm and it would have been nice if the paper had talked about other
algorithms. Also, there is a lack of experiments and theoretical results.
Future Work : I think some of the future work could be to try and extend
the various algorithms to improve their performance. The field seems to be
quite open to new ideas and algorithms.
This archive was generated by hypermail 2.1.6 : Tue May 13 2003 - 11:03:03 PDT