Title: Making Decisions with Spatially and Temporally Uncertain Data

Abstract: Taking advantage of known uncertainty patterns (e.g., uncertainty distributions) and strategically exploiting the uncertainty priors has been a prevalent methodology in the design of planning algorithms for autonomous systems that need to cope with various sources of uncertainty. However, in many scenarios the pattern of uncertainty is dynamic -- it reveals new patterns varying with some underlying parameters. We envision that such "multi-dimensional" uncertainty dynamics represent a more general description of the uncertainty, but could introduce extra complexity in modeling and solving planning problems. Therefore, we propose a principled planning and decision-making framework that takes into account multiple dimensional uncertainty.
One relevant and concrete example is to deal with uncertainty that varies significantly both spatially and temporally. To incorporate such variability, in this paper we propose to generalize Markov Decision Processes, by incorporating the time-varying uncertainty and redesigning the fundamental value propagation mechanisms. We transform this decision-making method that considers only immediate and static uncertainty descriptions to an upgraded planning framework that is able to adapt to future time-varying uncertainty over some horizon.