Enhancing Efficiency of Safe Reinforcement Learning via Sample Manipulation

TL;DR

This paper introduces ESPO, a novel safe reinforcement learning method that improves sample efficiency and safety by dynamically manipulating samples during training, with theoretical guarantees and superior empirical performance.

Contribution

The paper proposes ESPO, a new safe RL approach that adaptively adjusts sampling to enhance efficiency and safety, with proven convergence and better sample complexity.

Findings

ESPO outperforms baselines in reward maximization and safety constraints.

ESPO reduces sample requirements by 25-29% compared to existing methods.

ESPO decreases training time by 21-38%.

Abstract

Safe reinforcement learning (RL) is crucial for deploying RL agents in real-world applications, as it aims to maximize long-term rewards while satisfying safety constraints. However, safe RL often suffers from sample inefficiency, requiring extensive interactions with the environment to learn a safe policy. We propose Efficient Safe Policy Optimization (ESPO), a novel approach that enhances the efficiency of safe RL through sample manipulation. ESPO employs an optimization framework with three modes: maximizing rewards, minimizing costs, and balancing the trade-off between the two. By dynamically adjusting the sampling process based on the observed conflict between reward and safety gradients, ESPO theoretically guarantees convergence, optimization stability, and improved sample complexity bounds. Experiments on the Safety-MuJoCo and Omnisafe benchmarks demonstrate that ESPO significantly outperforms existing primal-based and primal-dual-based baselines in terms of reward maximization and constraint satisfaction. Moreover, ESPO achieves substantial gains in sample efficiency, requiring 25--29% fewer samples than baselines, and reduces training time by 21--38%.