Sonic Benchmark

Sonic Benchmark是一个基于Sonic the Hedgehog视频游戏系列的强化学习基准数据集，旨在评估转移学习和少量学习算法在强化学习领域的性能。该数据集包含58个保存状态，来自三个类似的游戏，每个保存状态代表游戏中的一个不同关卡。数据集设计用于测试任务间的泛化能力，特别是少量学习，通过提供多个相似任务的样本，为强化学习算法提供了一个合适的测试平台。该数据集的应用领域包括探索新环境的强化学习算法，以及通过转移学习提高学习速度的研究。

Sonic Benchmark is a reinforcement learning benchmark dataset based on the Sonic the Hedgehog video game series, which aims to evaluate the performance of transfer learning and few-shot learning algorithms in the reinforcement learning domain. This dataset contains 58 saved states from three similar games, with each saved state representing a distinct level within the games. The dataset is designed to test cross-task generalization capabilities, particularly for few-shot learning, by providing samples of multiple similar tasks, thus serving as a suitable testbed for reinforcement learning algorithms. The application areas of this dataset include reinforcement learning algorithms that explore new environments, as well as research aiming to improve learning efficiency via transfer learning.