DIAYN Skills Dataset

该数据集展示了通过DIAYN方法在各种模拟机器人任务中学习到的一系列技能，包括但不限于行走、跳跃和在不同环境中导航等多样化技能。此外，数据集还包含了关于这些技能的训练动态信息，以及预训练技能在下游任务中的有效性。这些技能是在不同任务中学习的，涵盖了从简单到复杂、自由度各异的环境。该数据集的任务是研究强化学习中的无监督技能发现。

This dataset showcases a collection of skills learned using the DIAYN method across a wide range of simulated robotic tasks, encompassing diverse capabilities such as walking, jumping, and navigation in varied environments. Additionally, the dataset includes training dynamics data for these skills, alongside the performance effectiveness of pre-trained skills when applied to downstream tasks. These skills are trained across distinct tasks, covering environments that span from simple to complex configurations with differing degrees of freedom. The core task of this dataset is to support research on unsupervised skill discovery in reinforcement learning.