Abstrakt
This article briefly introduces selected probabilistic approaches to search optimal behaviour of an agent in dynamic systems. It describes two possible extensions of Markov decision processes (MDP) which are the standard for probabilistic planning.
Our uncertainty about transition probabilities is described by Markov decision processes with imprecise probabilities (MDP-IP). Missing information about actual state can be handled by Partially observable Markov decision processes (POMDP).
This article compares both approaches and discusses their possible combination.