Avalon

该数据集名为Avalon，专为研究强化学习（RL）的泛化能力而设计。它包含了一系列在程序生成的3D世界中进行的任务，在这些任务中，智能体需要穿越地形、狩猎或收集食物，同时避免危险。Avalon数据集还提供了高效的模拟器、一系列基准算法库，以及针对数百小时人类表现评估的评分指标。该数据集的任务聚焦于强化学习的一般化能力。

This dataset, named Avalon, is specifically designed for research on the generalization capabilities of reinforcement learning (RL). It contains a series of tasks set in procedurally generated 3D worlds, where agents need to traverse terrain, hunt or collect food while avoiding hazards. The Avalon dataset also provides an efficient simulator, a suite of benchmark algorithm libraries, and evaluation metrics based on hundreds of hours of human performance assessment. The tasks of this dataset focus on the generalization capabilities of reinforcement learning.