RL Unplugged
收藏资源简介:
RL Unplugged 是一套用于离线强化学习的基准测试。 RL Unplugged 是围绕以下考虑设计的:为了便于使用,数据集提供了统一的 API,一旦建立了通用管道,从业者就可以轻松地处理套件中的所有数据。这是论文 RL Unplugged: Benchmarks for Offline Reinforcement Learning 随附的数据集。在这套基准测试中,作者试图关注以下问题: 高维动作空间,例如运动人形域,有 56 维动作。高维观察。部分可观察性,观察具有以自我为中心的视野。探索困难,使用最先进的算法和模仿为困难环境生成数据。现实世界的挑战。
RL Unplugged is a suite of benchmarks for offline reinforcement learning. It is designed with the following core considerations: to facilitate ease of use, the dataset provides a unified API. Once the general pipeline is established, practitioners can easily process all the data in the suite. This is the dataset accompanying the paper "RL Unplugged: Benchmarks for Offline Reinforcement Learning". In this benchmark suite, the authors focus on the following key issues: high-dimensional action spaces (e.g., the humanoid locomotion domain with 56-dimensional actions), high-dimensional observations, partial observability where observations have egocentric visual fields, exploration challenges—generating data for challenging environments using state-of-the-art algorithms and imitation learning—and real-world challenges.

- RL Unplugged数据集首次发表,由DeepMind团队在NeurIPS会议上正式发布,标志着强化学习领域数据集的新里程碑。
- RL Unplugged数据集首次应用于大规模强化学习任务,显著提升了模型训练效率和性能,成为研究者广泛使用的基准数据集。
- RL Unplugged数据集在多个国际顶级会议上被引用和讨论,进一步验证了其在强化学习研究中的重要性和影响力。
- 1RL Unplugged: Benchmarks for Offline Reinforcement LearningDeepMind · 2020年
- 2A Deeper Look at the RL Unplugged BenchmarkUniversity of California, Berkeley · 2021年
- 3Offline Reinforcement Learning: Tutorial, Review, and Perspectives on Open ProblemsUniversity of California, Berkeley · 2020年
- 4Benchmarking Batch Deep Reinforcement Learning AlgorithmsUniversity of California, Berkeley · 2020年
- 5Offline Reinforcement Learning with Implicit Q-LearningDeepMind · 2021年



