From: Sumit Sanghai (sanghai_at_cs.washington.edu)
Date: Tue Apr 22 2003 - 11:09:32 PDT
Planning with Incomplete Information as Heuristic Search in Belief Space
--- Blai Bonet and Hector Geffner
Summary : The paper presents the problems of planning with incomplete information and observations and solves them by reducing them to heuristic search problems
Ideas : Some important ideas are the formulation of the problems and classifying them as (i) classical (ii) conformant (iii) contingent with deterministic observations
(iv) contingent with probabilistic observations and reducing it to POMDP, and finally solving them using Real time dynamic programming.
They also find that in contingent planning, A* doesn't work that well as compared to RTDP
Flaws : The experiments didn't bring out the advantages and disadvantages of the different algorithms. Instead they just tried to compare the execution times of the various algorithms (specially in conformant planning).
Future Work : The obvious open problem is to look for better heuristics (in terms of performance and speed) specially for infinite domain and also study how these algorithms scale.
This archive was generated by hypermail 2.1.6 : Tue Apr 22 2003 - 11:09:34 PDT