Partially Observable Markov Decision Process Article Index for
Partially
Website Links For
Markov
 

Information About

Partially Observable Markov Decision Process




Currently, most POMDPs are computationally Intractable to solve exactly for optimal behavior, so computer scientists have developed methods that approximate solutions for POMDPs.

POMDPs can be used in solving simple Path Planning problems for mobile robots. This application is sometimes called the "Kidnapped Robot Problem," framed by imagining a robot was moved to an unknown location in a known environment and now must figure out where it is and find its way home. An exact solution to the POMDP will generate the series of actions that is most likely to get it home with the least cost.


DEFINITION

A Partially Observable Markov Decision Process (POMDP) is a tuple (S,A,O,P,R), where

  • S is the State space,

  • A is the action space,

  • O is the observation space.