Real-World Reinforcement Learning Suite
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/jqueeney/robust-safe-rl
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是一套带有安全约束的连续控制任务，旨在评估强化学习算法的鲁棒性和安全性。这些任务包括Cartpole Swingup、Walker Walk、Walker Run、Quadruped Walk和Quadruped Run，每个任务的时间跨度为1000步，并设有特定的安全预算限制。该数据集涵盖了3个领域内的5个受约束任务，其目标是研究在连续控制任务上的安全强化学习。
This dataset consists of continuous control tasks with safety constraints, which is designed to evaluate the robustness and safety of reinforcement learning algorithms. These tasks include Cartpole Swingup, Walker Walk, Walker Run, Quadruped Walk and Quadruped Run. Each task has a time horizon of 1000 steps and is subject to specific safety budget constraints. The dataset covers five constrained tasks across three domains, and its purpose is to support research on safe reinforcement learning in continuous control tasks.
提供机构：
Real-World Reinforcement Learning Suite
