Pascal Poupart, Aarti Malhotra, Pei Pei, Kee-Eung Kim, Bongseok Goh, and Michael Bowling. Approximate Linear Programming for Constrained Partially Observable Markov Decision Processes. In Proceedings of the Twenty-Ninth Conference on Artificial Intelligence (AAAI), 2015. To Appear
In many situations, it is desirable to optimize a sequenceof decisions by maximizing a primary objective while respectingsome constraints with respect to secondary objectives.Such problems can be naturally modeled as constrainedpartially observable Markov decision processes (CPOMDPs)when the environment is partially observable. In this work,we describe a technique based on approximate linear programmingto optimize policies in CPOMDPs. The optimizationis performed offline and produces a finite state controllerwith desirable performance guarantees. The approach outperformsa constrained version of point-based value iteration ona suite of benchmark problems.
@InProceedings(15aaai-cpomdp, Title = "Approximate Linear Programming for Constrained Partially Observable Markov Decision Processes", Author = "Pascal Poupart and Aarti Malhotra and Pei Pei and Kee-Eung Kim and Bongseok Goh and Michael Bowling", Booktitle = "Proceedings of the Twenty-Ninth Conference on Artificial Intelligence (AAAI)", Year = "2015", Note = "To Appear", AcceptRate = "27%", AcceptNumbers = "531 of 1991" )