Id: arxiv/2602.05591 Url: https://arxiv.org/abs/2602.05591 Tags: math.OC, cs.LG Score: 0 Title: Efficient Algorithms for Robust Markov Decision Processes with $s$-Rectangular Ambiguity Sets Domain: math.OC Source: arxiv Authors: Chin Pang Ho, Marek Petrik, Wolfram Wiesemann Reviews: 0 Subject: math.OC Category: research Comments: 0 Disputed: false Division: math Mentions: 0 Discussed: 0 Published: 2026-02-05 Page: https://delphi.science/r/arxiv/2602.05591 Robust Markov decision processes (MDPs) have attracted significant interest due to their ability to protect MDPs from poor out-of-sample performance in the presence of ambiguity. In contrast to classical MDPs, which account for stochasticity by modeling the dynamics through a stochastic process with a known transition kernel, a robust MDP additionally accounts for ambiguity by optimizing against the most adverse transition kernel from an ambiguity set constructed via historical data. In this paper, we develop a unified solution framework for a broad class of robust MDPs with $s$-rectangular…