Unsupervised Reinforcement Learning Benchmark (URLB)

URLB是由加州大学伯克利分校的Michael Laskin等人开发的一个用于评估无监督强化学习算法的数据集。该数据集包含12个连续控制任务，分为三个领域，旨在评估算法在无监督预训练后的适应效率。数据集基于DeepMind Control Suite，提供了统一的代码库和评估流程，以促进无监督强化学习算法的发展和比较。

URLB is a dataset developed by Michael Laskin et al. from the University of California, Berkeley, for evaluating unsupervised reinforcement learning algorithms. This dataset contains 12 continuous control tasks divided into three domains, aiming to evaluate the adaptation efficiency of algorithms after unsupervised pre-training. Based on the DeepMind Control Suite, the dataset provides a unified codebase and evaluation pipeline to facilitate the development and comparison of unsupervised reinforcement learning algorithms.