CoRL2026-CSI/SO101-CloseDrawer_Ours_50epi
收藏资源简介:
该数据集是一个用于机器人学研究的机器人操作数据集,具体任务是关闭抽屉。它使用LeRobot框架创建,涉及so101_follower机器人类型。数据集包含50个完整的情节(episodes),总计13143帧数据,以10帧每秒的速率采集。数据以parquet文件格式存储,并附有相应的视频文件。数据集的特征丰富,包括机器人的状态观测(如6个关节的位置、末端执行器的XYZRPY坐标、夹爪的二进制状态)、动作(6个关节的目标位置)、来自顶部和左手腕摄像头的图像观测(分辨率为480x640的RGB视频)、技能信息(如自然语言描述、验证问题、技能类型、进度、目标关节位置和末端执行器位置)、子任务信息(如自然语言描述、对象名称、目标位置)以及时间戳、帧索引、情节索引等元数据。数据集仅包含训练集,适用于机器人控制、模仿学习或强化学习等任务。
This dataset is a robotics manipulation dataset for research purposes, specifically focused on the task of closing a drawer. It was created using the LeRobot framework and involves the so101_follower robot type. The dataset contains 50 complete episodes, totaling 13143 frames, collected at a rate of 10 frames per second. Data is stored in parquet format along with corresponding video files. The dataset features a wide range of information, including robot state observations (such as positions of 6 joints, XYZRPY coordinates of the end-effector, binary gripper state), actions (target positions for 6 joints), image observations from top and left wrist cameras (RGB videos with 480x640 resolution), skill information (e.g., natural language descriptions, verification questions, skill type, progress, target joint positions, and end-effector positions), subtask information (e.g., natural language descriptions, object names, target positions), and metadata like timestamps, frame indices, and episode indices. The dataset includes only a training split and is suitable for tasks such as robot control, imitation learning, or reinforcement learning.




