Balance Reward and Safety Optimization for Safe Reinforcement Learning: A Perspective of Gradient Manipulation

arXiv:2405.01677v1 Announce Type: new
Abstract: Ensuring the safety of Reinforcement Learning (RL) is crucial for its deployment in real-world applications. Nevertheless, managing the trade-off between reward and safety during exploration presents a significant challenge. Improving reward performance through policy adjustments may adversely affect safety performance. In this study, we aim to address this conflicting relation by leveraging the theory of gradient manipulation. Initially, we analyze the conflict between reward and safety gradients. Subsequently, we tackle the balance between reward and …

abstract applications arxiv balance challenge cs.ai cs.lg deployment exploration gradient improving manipulation optimization performance perspective policy reinforcement reinforcement learning safe safety through trade trade-off type world