Adaptive Primal-Dual Method for Safe Reinforcement Learning
This work addresses the problem of stable and efficient constrained policy optimization for researchers and practitioners in Safe Reinforcement Learning, representing an incremental improvement over existing methods.
The paper tackles the challenge of applying primal-dual methods to Safe Reinforcement Learning by proposing an adaptive primal-dual algorithm that adjusts learning rates to Lagrangian multipliers, showing it outperforms or matches constant learning rate methods with more stable training in experiments across four environments.
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.