so101_hamburger_to_paper_bag_merged
收藏资源简介:
该数据集是一个机器人操作数据集,使用LeRobot工具创建,专门针对双臂机器人(bi_so_follower类型)执行任务的数据收集。数据集包含50个episodes和37297帧,帧率为25fps,特征包括动作和观察:动作涉及12个关节位置(例如左肩平移、左肩抬升、左肘弯曲等),观察则包括机器人状态(同样为12个关节位置)以及来自多个摄像头视角的图像视频(如左腕左视图、左顶部视图和右腕右视图),视频分辨率为480x640,编码为av1。数据以parquet格式存储,并附带时间戳、帧索引、episode索引等信息,适用于机器人学习和控制任务。
This dataset is a robot manipulation dataset created using the LeRobot tool, specifically designed for data collection from a dual-arm robot (bi_so_follower type) performing tasks. It contains 50 episodes and 37297 frames with a frame rate of 25fps, featuring actions and observations: actions involve 12 joint positions (e.g., left shoulder translation, left shoulder elevation, left elbow flexion, etc.), while observations include robot states (also 12 joint positions) and image videos from multiple camera perspectives (such as left wrist left view, left top view, and right wrist right view), with video resolution of 480x640 and av1 encoding. The data is stored in parquet format and includes timestamps, frame indices, episode indices, etc., making it suitable for robot learning and control tasks.
数据集概要
- 数据集名称:so101_hamburger_to_paper_bag_merged
- 许可证:Apache-2.0
- 任务类别:机器人学(robotics)
- 标签:LeRobot
- 创建工具:LeRobot
数据集结构
- 数据文件格式:Parquet(
data/*/*.parquet) - 视频编码:AV1,分辨率 480x640,帧率 25 FPS
- 总剧集数:50
- 总帧数:37,297
- 总任务数:1
- 机器人类型:bi_so_follower(双臂机器人)
特征字段
| 特征名称 | 数据类型 | 形状/描述 |
|---|---|---|
| action | float32 | 12维,包含左右双臂各6个关节位置(肩、肘、腕、夹爪) |
| observation.state | float32 | 12维,与action结构相同 |
| observation.images.left_left_wrist | 视频 | 480x640x3,左腕摄像头 |
| observation.images.left_top | 视频 | 480x640x3,左上方摄像头 |
| observation.images.right_right_wrist | 视频 | 480x640x3,右腕摄像头 |
| timestamp | float32 | 1维,时间戳 |
| frame_index | int64 | 1维,帧索引 |
| episode_index | int64 | 1维,剧集索引 |
| index | int64 | 1维,全局索引 |
| task_index | int64 | 1维,任务索引(仅1个任务) |
数据分割
- 训练集:包含全部50个剧集(索引0至49)
- 数据文件大小:100 MB
- 视频文件大小:200 MB
其他信息
- 数据集主页:未提供
- 相关论文:未提供
- 引用格式:未提供



