oxe-auge/fractal20220817_data_train_35000_37500_augmented

该数据集名为fractal20220817_data_train_35000_37500_augmented，是一个用于机器人增强学习的大规模数据集。数据集包含多种机器人（如jaco、kinova3、kuka_iiwa等）的图像和状态信息，旨在支持跨实体策略学习的扩展。数据集提供了丰富的特征，包括原始机器人和增强后机器人的图像、末端执行器位置、关节角度等。此外，数据集还包含自然语言指令，用于描述任务。数据集的总帧数为108,469，帧率为3fps，总共有2,500个episode。数据以parquet格式存储，并提供了详细的数据布局和特征描述。

The dataset named fractal20220817_data_train_35000_37500_augmented is a large-scale dataset for robotic reinforcement learning. It includes images and state information of various robots (e.g., jaco, kinova3, kuka_iiwa) to support scaling cross-embodiment policy learning. The dataset provides rich features, including images of original and augmented robots, end-effector positions, joint angles, etc. Additionally, the dataset contains natural language instructions to describe tasks. The total number of frames is 108,469, with a frame rate of 3fps, and there are 2,500 episodes in total. The data is stored in parquet format, with detailed data layout and feature descriptions provided.