arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2602.15076cs.LGstat.ML
收藏

在线约束马尔可夫决策过程的近最优样本复杂度

Near-Optimal Sample Complexity for Online Constrained MDPs

University of California, Los Angeles(加州大学洛杉矶分校)

Chang Liu, Yunfan Li, Lin F. Yang

AI总结:

本文提出了一种基于模型的对偶算法,解决了在线约束马尔可夫决策过程的安全性问题,证明了在允许小规模违规的情况下,算法能以近最优样本复杂度生成安全策略。

摘要:

安全是强化学习(RL)中的一个基本挑战,特别是在自动驾驶、机器人和医疗等现实应用中。为了解决这个问题,约束马尔可夫决策过程(CMDPs)被广泛用于在优化性能的同时强制安全约束。然而,现有方法往往遭受显著的安全违规或需要高样本复杂度来生成近最优策略。我们考虑两种设置:放松可行性,允许小规模违规,和严格可行性,不允许任何违规。我们提出了一种基于模型的对偶算法,该算法平衡了后悔和有界约束违规,借鉴了在线RL和约束优化的技术。对于放松可行性,我们证明我们的算法在任意高概率下返回一个ε-最优策略,具有ε-有界违规,需要~O(SA H³ / ε²)次学习回合,与无约束MDPs的下界相匹配。对于严格可行性,我们证明我们的算法在任意高概率下返回一个ε-最优策略,零违规,需要~O(SA H⁵ / (ε² ζ²))次学习回合,其中ζ是问题相关的Slater常数,表征可行区域的大小。这一结果与学习CMDPs时可访问生成模型的下界相匹配。我们的结果表明,在在线设置下学习CMDPs与学习生成模型一样容易,并且当允许小规模违规时,其难度不比学习无约束MDPs更大。

英文摘要:

Safety is a fundamental challenge in reinforcement learning (RL), particularly in real-world applications such as autonomous driving, robotics, and healthcare. To address this, Constrained Markov Decision Processes (CMDPs) are commonly used to enforce safety constraints while optimizing performance. However, existing methods often suffer from significant safety violations or require a high sample complexity to generate near-optimal policies. We address two settings: relaxed feasibility, where small violations are allowed, and strict feasibility, where no violation is allowed. We propose a model-based primal-dual algorithm that balances regret and bounded constraint violations, drawing on techniques from online RL and constrained optimization. For relaxed feasibility, we prove that our algorithm returns an $\varepsilon$-optimal policy with $\varepsilon$-bounded violation with arbitrarily high probability, requiring $\tilde{O}\left(\frac{SAH^3}{\varepsilon^2}\right)$ learning episodes, matching the lower bound for unconstrained MDPs. For strict feasibility, we prove that our algorithm returns an $\varepsilon$-optimal policy with zero violation with arbitrarily high probability, requiring $\tilde{O}\left(\frac{SAH^5}{\varepsilon^2ζ^2}\right)$ learning episodes, where $ζ$ is the problem-dependent Slater constant characterizing the size of the feasible region. This result matches the lower bound for learning CMDPs with access to a generative model. Our results demonstrate that learning CMDPs in an online setting is as easy as learning with a generative model and is no more challenging than learning unconstrained MDPs when small violations are allowed.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑