Partially Observable Markov Decision Processes in Robotics: A Survey

Noisy sensing, imperfect control, and environment changes are defining characteristics of many real-world robot tasks. The partially observable Markov decision process (POMDP) provides a principled mathematical framework for modeling solving control tasks under uncertainty. Over the last decade, it has seen successful applications, spanning localization navigation, search tracking, autonomous driving, multi-robot systems, manipulation, human-robot interaction. This survey aims to bridge gap between development POMDP models algorithms at one end application diverse other. It analyzes these connects them with algorithmic properties effective solution. For practitioners, some key task in deciding when how apply POMDPs successfully. algorithm designers, new insights into unique challenges applying systems points promising directions further research.