Journal of Arti cial Intelligence Research Submitted published
Nonapproximability Results for
Partially Observable Markov Decision Pro cesses
Christopher Lusena lusena cs engr uky edu
University of Kentucky
Dept of Computer Science Lexington KY
Judy Goldsmith goldsmit cs engr uky edu
University of Kentucky
Dept of Computer Science Lexington KY
Martin Mundhenk mundhenk ti uni trier de
Universit at Trier
FB IV Informatik D Trier Germany
Abstract
We show that for several variations of partially observable Markov decision pro cesses
p olynomial time algorithms for nding control p olicies are unlikely to or simply don t have
guarantees of nding p olicies within a constant factor or a constant summand of optimal
Here unlikely means unless some complexity classes collapse where the collapses con
sidered are P NP P PSPACE or P EXP Until or unless these collapses are shown
to hold any control policy designer must choose b etween such p erformance guarantees and
e cient computation