Continuous Arcade Learning Environment (CALE)

该数据集名为CALE，是 Arcade Learning Environment（ALE）的扩展版本，支持连续动作，这使得在Atari 2600游戏系统中可以对连续控制代理和价值基础代理进行基准测试和评估。该数据集提供了使用Soft Actor-Critic代理的初步基准结果，并且评估协议包括了粘性行动和跳帧。在规模上，该数据集包含了60款游戏，总计2亿帧，以及26款游戏用于Atari 100k基准测试。其任务是对连续控制代理进行基准测试和评估。

This dataset, named CALE, is an extended version of the Arcade Learning Environment (ALE) that supports continuous actions, enabling benchmarking and evaluation of both continuous-control agents and value-based agents on the Atari 2600 game system. It provides preliminary benchmark results using the Soft Actor-Critic agent, and its evaluation protocol includes sticky actions and frame skip. In terms of scale, the dataset covers 60 games with a total of 200 million frames, and includes 26 games for the Atari 100k benchmark. The core task of this dataset is to benchmark and evaluate continuous-control agents.