{"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","summary":"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…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-02-05","reactions":{}}