URLB (Unsupervised Reinforcement Learning Benchmark)

URLB 由两个阶段组成：无奖励预训练和具有外部奖励的下游任务适应。它以 DeepMind Control Suite 为基础，提供来自三个域的十二个连续控制任务进行评估。

URLB consists of two stages: reward-free pretraining and downstream task adaptation with external rewards. Based on the DeepMind Control Suite, it provides twelve continuous control tasks from three domains for evaluation.

搜集汇总

背景与挑战

背景概述

URLB是一个无监督强化学习基准，包含无奖励预训练和下游任务适应两个阶段。它基于DeepMind Control Suite，提供来自三个领域的十二个连续控制任务用于评估。