Policy Optimization Achieves Data-Dependent Regret Bounds in MDPs with Unknown Transitions
Abstract
We study policy optimization for online episodic tabular Markov decision processes with unknown transition kernels, aiming for best-of-both-worlds guarantees together with data-dependent regret bounds. Recent work (Dann et al., 2023; Li et al., 2026) has shown that policy optimization can adapt to both adversarial and stochastic losses with first-order, second-order, and path-length bounds, but only under known transitions, leaving open whether such data-dependent guarantees are achievable by policy optimization when the transition kernel is unknown. We resolve this by developing a new algorithm based on optimistic follow-the-regularized-leader that attains these guarantees under unknown transitions. The key ingredient is a new design of optimistic -function estimators together with a data-dependent transition bonus that controls estimator bias through the loss-prediction error. Our analysis further identifies an unavoidable transition-dependent complexity term that captures the intrinsic cost of estimating the transition kernel. As a result, we obtain first-order, second-order, and path-length bounds with the transition-dependent complexity term while simultaneously achieving gap-dependent regret in the stochastic regime.
Cite
@article{arxiv.2606.31769,
title = {Policy Optimization Achieves Data-Dependent Regret Bounds in MDPs with Unknown Transitions},
author = {Mingyi Li and Taira Tsuchiya and Kenji Yamanishi},
journal= {arXiv preprint arXiv:2606.31769},
year = {2026}
}
Comments
70 pages, 2 tables