Solving POMDPs by Searching the Space of Finite Policies

by: Nicolas Meuleau, Kee-Eung Kim, Leslie P Kaelbling, Anthony R Cassandra

pp. 417-426.

View FullText article

CiteSeer (at Penn State), CiteSeer (at MIT), CiteSeer (at Zurich), CiteSeer (at Singapore)

Reviews [Write a review of this article]

There are no reviews of this article

Notes for this article

noel has 0 private notes и ещё 1 public note for this article.

Model-based method for learning finite-state controllers for POMDPs. Presents two algorithms, one a globally optimal branch-and-bound search, and the other a locally optimal gradient ascent method.

noel (public ) - 2005-04-04 12:46:43

Find related articles from these CiteULike users

noel, szityu, bellatrix, shibu184, eowyn

Find related articles with these CiteULike tags

policy, policy-based, pomdp, rl, search

Abstract

Solving partially observable Markov decision processes (POMDPs) is highly intractable in general, at least in part because the optimal policy may be infinitely large. In this paper, we explore the problem of finding the optimal policy from a restricted set of policies, represented as finite state automata of a given size. This problem is also intractable, but we show that the complexity can be greatly reduced when the POMDP and/or policy are further constrained. We demonstrate good...

@inproceedings{citeulike:144031, abstract = {Solving partially observable Markov decision processes (POMDPs) is highly intractable in general, at least in part because the optimal policy may be infinitely large. In this paper, we explore the problem of finding the optimal policy from a restricted set of policies, represented as finite state automata of a given size. This problem is also intractable, but we show that the complexity can be greatly reduced when the POMDP and/or policy are further constrained. We demonstrate good...}, author = {Meuleau, Nicolas and Kim, Kee-Eung and Kaelbling, Leslie P. and Cassandra, Anthony R. }, citeulike-article-id = {144031}, comment = {Model-based method for learning finite-state controllers for POMDPs. Presents two algorithms, one a globally optimal branch-and-bound search, and the other a locally optimal gradient ascent method.}, keywords = {policy-based, pomdp}, pages = {417--426}, posted-at = {2005-04-01 10:04:12}, priority = {0}, title = {Solving POMDPs by Searching the Space of Finite Policies}, url = {http://citeseer.ist.psu.edu/meuleau99solving.html} }

RIS record

TY - CONF ID - citeulike:144031 L3 - citeulike-article-id:144031 TI - Solving POMDPs by Searching the Space of Finite Policies SP - 417 EP - 426 N2 - Solving partially observable Markov decision processes (POMDPs) is highly intractable in general, at least in part because the optimal policy may be infinitely large. In this paper, we explore the problem of finding the optimal policy from a restricted set of policies, represented as finite state automata of a given size. This problem is also intractable, but we show that the complexity can be greatly reduced when the POMDP and/or policy are further constrained. We demonstrate good... KW - policy-based KW - pomdp N1 - Model-based method for learning finite-state controllers for POMDPs. Presents two algorithms, one a globally optimal branch-and-bound search, and the other a locally optimal gradient ascent method. AU - Meuleau, Nicolas AU - Kim, Kee-Eung AU - Kaelbling, Leslie AU - Cassandra, Anthony UR - http://citeseer.ist.psu.edu/meuleau99solving.html ER -

RIS BibTeX