HyeonseokE/SO101-OpenDrawer_Ours_40epi
收藏资源简介:
该数据集是一个机器人操作数据集,使用LeRobot框架创建,专门用于训练和评估机器人学习模型。数据集包含40个训练片段,总计11191帧,数据以parquet文件格式存储,视频以mp4格式存储,帧率为10fps。数据集记录了机器人(型号为so101_follower)在执行任务过程中的状态信息,包括6个关节的位置观测、末端执行器的位姿(XYZRPY坐标)、夹爪的二进制状态,以及来自顶部和左腕摄像头的视频观测(分辨率640x480)。动作空间为6维关节位置控制。此外,数据集还包含技能级别的自然语言描述、验证问题、技能类型、进度和目标位置(关节空间和笛卡尔空间),以及子任务的自然语言描述、对象名称和目标位置。数据集适用于机器人模仿学习、强化学习或视觉语言任务的研究。
This dataset is a robotic manipulation dataset created using the LeRobot framework, designed for training and evaluating robot learning models. It contains 40 training episodes with a total of 11,191 frames, stored in parquet format for data and mp4 format for videos at 10 fps. The dataset records state information of a robot (model so101_follower) during task execution, including observations of 6 joint positions, end-effector pose (XYZRPY coordinates), gripper binary state, and video observations from top and left wrist cameras (resolution 640x480). The action space is 6-dimensional joint position control. Additionally, the dataset includes skill-level natural language descriptions, verification questions, skill types, progress, and goal positions (in joint and Cartesian space), as well as subtask natural language descriptions, object names, and target positions. It is suitable for research in robot imitation learning, reinforcement learning, or vision-language tasks.




