Robust Sequential Decision Making

We consider planning problems where the parameters of the problems are not known. The robust approach to sequential decision making is to assume that the worst possible realization within a predefined uncertainty set will occur at every stage. While this approach is tractable, its pessimistic nature may lead to extremely conservative solutions. We will discuss several approaches that work-around the inherent conservativeness of the standard robust approach while remaining tractable. The proposed approaches also offer interesting probabilistic guarantees on the performance of the computed policy under a probabilistic deviation model.

Authors

• Shie Mannor Technion - Israel Institute of Technology NVIDIA Research

Keywords

No keywords are indexed for this paper.