arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2402.00355cs.LGcs.AImath.OC
收藏

面向安全强化学习的自适应原始-对偶方法

Adaptive Primal-Dual Method for Safe Reinforcement Learning

Rensselaer Polytechnic Institute(伦斯勒理工学院);IBM T.J. Watson Research Center(IBM TJ·沃森研究中心);IBM Research(IBM研究院)

Weiqin Chen, James Onyejizu, Long Vu, Lan Hoang, Dharmashankar Subramanian, Koushik Kar, Sandipan Mishra, Santiago Paternain

AI总结:

针对安全强化学习中原始-对偶方法的学习率与拉格朗日乘子相互依赖的难题,提出自适应原始-对偶方法,理论证明其收敛性等特性,实验显示其性能更优、训练更稳定且鲁棒性好。

摘要:

原始-对偶方法在被建模为约束策略优化问题的安全强化学习(Safe Reinforcement Learning, SRL)中有着天然应用。但在实践中,由于每次求解内嵌的无约束强化学习问题时,学习率(learning rate, LR)与拉格朗日乘子(对偶变量)存在相互依赖关系,将原始-对偶方法应用于SRL颇具挑战。本文提出、分析并评估了面向SRL的自适应原始-对偶(adaptive primal-dual, APD)方法,该方法根据拉格朗日乘子调整两个自适应学习率,以在每次迭代中优化策略。我们从理论上证明了APD算法的收敛性、最优性和可行性。最后,我们在Bullet-Safey-Gym的四个经典环境中,采用PPO-Lagrangian和DDPG-Lagrangian两种当前最优的SRL算法,对实用型APD算法进行了数值评估。所有实验均表明,实用型APD算法性能优于(或持平于)固定学习率的情形,且训练过程更稳定。此外,我们通过实验证据验证了选取这两个自适应学习率的鲁棒性。

英文摘要:

Primal-dual methods have a natural application in Safe Reinforcement Learning (SRL), posed as a constrained policy optimization problem. In practice however, applying primal-dual methods to SRL is challenging, due to the inter-dependency of the learning rate (LR) and Lagrangian multipliers (dual variables) each time an embedded unconstrained RL problem is solved. In this paper, we propose, analyze and evaluate adaptive primal-dual (APD) methods for SRL, where two adaptive LRs are adjusted to the Lagrangian multipliers so as to optimize the policy in each iteration. We theoretically establish the convergence, optimality and feasibility of the APD algorithm. Finally, we conduct numerical evaluation of the practical APD algorithm with four well-known environments in Bullet-Safey-Gym employing two state-of-the-art SRL algorithms: PPO-Lagrangian and DDPG-Lagrangian. All experiments show that the practical APD algorithm outperforms (or achieves comparable performance) and attains more stable training than the constant LR cases. Additionally, we substantiate the robustness of selecting the two adaptive LRs by empirical evidence.

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。

↑