Safety-Gymnasium
收藏资源简介:
Safety-Gymnasium是由北京大学人工智能研究所开发的一个统一的安全强化学习基准环境。该数据集包含单代理和多代理场景下的安全关键任务,支持向量和视觉输入。此外,还提供了一个名为Safe Policy Optimization(SafePO)的算法库,包含16种最先进的SafeRL算法。Safety-Gymnasium旨在促进安全性能的评估和比较,推动强化学习在更安全、可靠和负责任的实际应用中的发展。数据集适用于评估和比较安全性能,解决复杂和高风险领域的智能系统可靠操作问题。
Safety-Gymnasium is a unified safety reinforcement learning benchmark environment developed by the Institute of Artificial Intelligence at Peking University. This dataset covers safety-critical tasks in both single-agent and multi-agent scenarios, supporting both vector and visual inputs. Furthermore, an algorithm library named Safe Policy Optimization (SafePO) is provided, which incorporates 16 state-of-the-art SafeRL algorithms. Safety-Gymnasium aims to facilitate the evaluation and comparison of safety performance, and promote the advancement of reinforcement learning for safer, more reliable and responsible real-world applications. This dataset is suitable for evaluating and comparing safety performance, and addressing the challenge of reliable operation of intelligent systems in complex and high-risk domains.

- 1Safety-Gymnasium: A Unified Safe Reinforcement Learning Benchmark北京大学人工智能研究所 · 2023年



