Tighter complexity bounds and broader strongly polynomial uncertainty classes

Obtain tighter complexity bounds for robust policy iteration and identify broader uncertainty classes for robust Markov decision processes that admit strongly polynomial algorithms.

Background

The paper establishes strongly polynomial bounds at fixed discount for several uncertainty models, including $\ell_1$, interval, weighted $\ell_1$, and Wasserstein uncertainty, under specific structural analyses. It explicitly leaves unresolved the improvement of these bounds and the extension of strongly polynomial solvability to uncertainty classes beyond those treated in the paper.