clean_up_table_v2_final
收藏Hugging Face2026-05-24 更新2026-05-24 收录
下载链接:
https://huggingface.co/datasets/YOLO2431/clean_up_table_v2_final
下载链接
链接失效反馈官方服务:
资源简介:
该数据集使用LeRobot工具创建,是一个机器人学领域的数据集,专门针对雅马哈双手机器人(yam_bimanual)。它包含300个完整的情节,总计246,897帧数据,覆盖4个不同任务。数据以30fps的帧率采集,包含900个视频文件。数据结构包括观测状态(observation.state)和动作(action),均为32维浮点数组,具体特征包括左右机械臂的3D位置坐标、6D旋转表示、夹持器状态以及6个关节角度。此外,数据集提供三个视角的视频观测:头部摄像头、左手腕摄像头和右手腕摄像头,每个视频的分辨率为640x360,3通道彩色,编码为h264。元数据包括时间戳、帧索引、情节索引和任务索引。数据集仅包含训练集(train split),所有300个情节均用于训练。数据以parquet格式分块存储,每个块包含1000帧。
This dataset is created using LeRobot and is a robotics dataset specifically for the Yamaha bimanual robot (yam_bimanual). It contains 300 complete episodes, totaling 246,897 frames, covering 4 different tasks. The data is collected at 30fps and includes 900 video files. The data structure consists of observation.state and action, both as 32-dimensional floating-point arrays, with specific features including 3D position coordinates, 6D rotation representation, gripper status, and 6 joint angles for both left and right arms. Additionally, the dataset provides video observations from three perspectives: head camera, left wrist camera, and right wrist camera, each with a resolution of 640x360, 3-channel color, encoded as h264. Metadata includes timestamps, frame indices, episode indices, and task indices. The dataset only includes a training split (train split), with all 300 episodes used for training. The data is stored in parquet format in chunks, with each chunk containing 1000 frames.
提供机构:
YOLO2431
创建时间:
2026-05-15
原始信息汇总
数据集概述
- 名称:clean_up_table_v2_final
- 许可证:Apache-2.0
- 任务类型:机器人学(Robotics)
- 创建工具:LeRobot
数据集规模
- 总片段数(Episodes):300
- 总帧数(Frames):246,897
- 总任务数(Tasks):4
- 总视频数(Videos):900
- 帧率(FPS):30
数据集结构
- 机器人类型:yam_bimanual(双臂机器人)
- 数据路径格式:
data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet - 视频路径格式:
videos/chunk-{episode_chunk:03d}/{video_key}/episode_{episode_index:06d}.mp4 - 划分:全部数据(片段索引0-299)用于训练(train)
特征详情
观测状态(observation.state)
- 数据类型:float32
- 形状:[32]
- 包含变量:左右臂的位置(x, y, z)、旋转6D表示(6个值)、夹爪状态(1个值)以及6个关节角度(各6个值),共计32维。
动作(action)
- 数据类型:float32
- 形状:[32]
- 包含变量:与观测状态完全一致,共计32维。
图像观测(observation.images)
- 三个摄像头视角:
- head(头部摄像头)
- left_wrist(左腕摄像头)
- right_wrist(右腕摄像头)
- 每个视角:
- 数据类型:视频(video)
- 形状:[3, 360, 640](通道数3,高度360,宽度640)
- 视频编码:h264,像素格式yuv420p,帧率30 FPS,无音频。
其他特征
- timestamp:时间戳(float32,形状[1])
- frame_index:帧索引(int64,形状[1])
- episode_index:片段索引(int64,形状[1])
- index:全局索引(int64,形状[1])
- task_index:任务索引(int64,形状[1])



