DSRL
收藏资源简介:
DSRL数据集是由卡内基梅隆大学创建的,用于离线安全强化学习研究的全面基准套件。数据集包含75000条轨迹,涵盖了从机器人控制到自动驾驶的38个流行安全强化学习任务。通过先进的强化学习算法,数据集的收集过程采用了系统的方法,包括数据后处理过滤器,以模拟各种数据收集条件。此外,数据集还提供了优雅且可扩展的流行离线安全强化学习算法实现,以加速该领域的研究。DSRL数据集旨在为研究人员和实践者提供一个标准测试平台,以评估和比较安全学习算法,促进更稳健和可靠的离线安全强化学习解决方案的发展。
The DSRL Dataset is a comprehensive benchmark suite developed by Carnegie Mellon University for offline safe reinforcement learning research. It contains 75,000 trajectories, covering 38 popular safe reinforcement learning tasks ranging from robotic control to autonomous driving. The dataset collection employed a systematic methodology leveraging advanced reinforcement learning algorithms, and incorporated data post-processing filters to simulate diverse data collection conditions. Furthermore, the dataset also provides elegant and scalable implementations of popular offline safe reinforcement learning algorithms to accelerate research in this field. The DSRL Dataset aims to provide researchers and practitioners with a standard testbed for evaluating and comparing safe learning algorithms, thereby fostering the development of more robust and reliable offline safe reinforcement learning solutions.




