Distracting DeepMind Control Suite
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/DMU-XMU/DSR
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是一个用于评估视觉强化学习算法的具有挑战性的基准，它包含了具有分散注意力的视频背景的连续控制任务。此外，该数据集用来自DAVIS 2017数据集的自然视频替换了干净的背景，以测试视觉任务中状态表征和场景泛化的性能。该数据集涵盖了六个连续控制任务，旨在评估视觉强化学习任务中的表征能力和策略性能。
This dataset is a challenging benchmark for evaluating visual reinforcement learning algorithms, which includes continuous control tasks with distracting video backgrounds. Furthermore, it replaces the clean backgrounds with natural videos sourced from the DAVIS 2017 dataset to test the performance of state representation and scene generalization in visual tasks. This dataset encompasses six continuous control tasks designed to evaluate the representation capability and policy performance in visual reinforcement learning tasks.
提供机构：
DeepMind
