Control theoretic tools for robust decision planning

Abstract

Continuous planning is emerging as the leading solution to the problem of planning for systems with significant uncertainty. However, continuous planning can yield unexpected poor performance due to the necessary incompleteness of the system models used for planning. In this paper, we show how the ideas of plan repair and discount factors can be formalized, quantified, and extended to develop robust continuous planners. This is achieved without increasing computational complexity or requiring extensive additional modeling.