Compiling POMDP models for a multimodal service robot from background knowledge

Abstract

This paper presents an approach to create POMDP models, used for decision making by an autonomous service robot, from background knowledge. This allows the power of POMDP decision making to be applied on multimodal service robots with quite distinct stochastic dynamics in different modalities and a generally high model complexity. The two tiered approach presented allows both fine grained model adaptation as well as semantically more transparent knowledge representation and easy composition of new scenarios. The application of the process on a realistic mission scenario performed by a physical service robot is presented as an example. © 2008 Springer-Verlag Berlin Heidelberg.