部分可观测马尔可夫决策过程算法综述

部分可观测马尔可夫决策过程(POMDP)是马尔可夫决策过程(MDP)的扩展,它允许系统的状态信息部分可知。但POMDP的可能应用大部分没有实现,这主要是因为缺乏有效的算法。POMDP的算法分为近似算法和精确算法,精确算法是构造近似算法的基础。介绍了POMDP模型后,对离散时间、有限状态集的POMDP精确算法和近似算法进行了综述,分析了造成POMDP难以求解的主要原因,提出了进一步的研究方向。

Survey of algorithms for partially observable Markov decision processes

A partially observable Markov decision process(POMDP) is an extension of a Markov decision process(MDP),which can partially keep the state of the system under observation.The applied potential for POMDP remains largely unrealized due to lack of tractable solution methodologies.The POMDP algorithms can divide into the approximate algorithms and the exact algorithms,and the exact algorithms are the base of the approximate algorithms.The exact and approximate algorithms for solving discrete-time,finite POMDP over finite horizon are summarized.In the end the reasons why POMDP problems are intractable and the future research directions are proposed.