EpicPinkPenguin/visual_dm_control

Visual DeepMind Control数据集包含了在DeepMind控制套件中的20个环境中由Dreamer V3强化学习智能体生成的专家轨迹。与其它数据集不同，该数据集提供了图像观察结果而不是状态。每个数据实例代表一个步骤，包括观察、动作、奖励以及是否终止或截断的信息。数据集分为训练集（90%）和测试集（10%），每个环境的数据集总共包含1000万个步骤。

The Visual DeepMind Control dataset contains expert trajectories generated by a Dreamer V3 reinforcement learning agent trained on various environments in the DeepMind Control Suite. Unlike other datasets, this dataset features image observations instead of states. Each data instance represents a single step with information such as observation, action, reward, and episode status. The dataset is split into training (90%) and testing (10%) subsets, with a total of 10 million steps across all environments.