Survey article on POMDPs in robotics in IEEE T-RO

Partially observable Markov decision processes (POMDPs) are a mathematical model for decision-making under uncertainty. Among other things, POMDPs can help robots to reliably handle noisy sensing, imperfect control, and environment changes. Together with my co-authors David Hsu and Joni Pajarinen, we have prepared a survey article that outlines how POMDPs have been applied in robot applications such as navigation, autonomous driving, human-robot coordination, or multi-robot coordination.

Preparing the survey was a really interesting project that allowed us to look at the developments over the last 10-20 years very broadly, and to distill it into a report that we hope will be useful for both roboticists and algorithm designers. For robotics practioners, we hope the survey provides guidance on when and how to apply POMDPs to robot tasks successfully. For POMDP algorithm designers, the survey highlights challenges of applying POMDPs to robot systems and points to promising new directions for further research.

The survey is due to appear in IEEE Transactions on Robotics (T-RO), and an author version of the accepted manuscript is available already now on arXiv!