fold_towel_merged_all
收藏资源简介：
该数据集使用LeRobot创建，是一个机器人学数据集，专注于双机械臂折叠毛巾任务。数据集包含271个回合、261,134帧数据，帧率为30fps。特征包括16维动作向量（控制左右机械臂的7个关节和夹爪位置）、16维观测状态向量（与动作相同）、以及三个视角的图像观测（顶部、右腕部、左腕部），图像分辨率为480x640，格式为视频。数据以parquet文件和视频文件存储，总数据文件大小为100MB，视频文件大小为200MB。机器人类型为bi_openarm_follower，用于训练和验证机器人控制模型。
This dataset was developed using LeRobot, and it is a robotics dataset dedicated to the dual-arm towel folding task. It encompasses 271 episodes and 261,134 frames of data, with a frame rate of 30 fps. The dataset features include a 16-dimensional action vector (controlling the 7 joints and gripper position of the left and right robotic arms respectively), a 16-dimensional observation state vector (consistent with the action vector), as well as image observations from three perspectives: top view, right wrist view, and left wrist view, with an image resolution of 480×640 and stored in video format. The data is saved in parquet files and video files, with the total size of all data files being 100 MB, and the video files alone occupying 200 MB. The robotic platform used is bi_openarm_follower, which is intended for training and validating robotic control models.
数据集概述
- 名称: fold_towel_merged_all
- 许可证: Apache-2.0
- 任务类别: 机器人技术 (robotics)
- 创建工具: 使用 LeRobot 创建
- 数据集大小: Parquet 数据文件约 100 MB，视频文件约 200 MB
数据集结构
- 总片段数 (episodes): 271
- 总帧数 (frames): 261,134
- 总任务数 (tasks): 1
- 帧率 (FPS): 30
- 机器人类型: bi_openarm_follower
- 数据分割: 训练集 (train) 包含全部 271 个片段（索引 0 至 270）
特征 (Features)
| 特征名称 | 数据类型 | 形状 | 说明 |
|---|---|---|---|
| action | float32 | [16] | 16维动作指令，包含左右臂各7个关节位置和1个夹爪位置 |
| observation.state | float32 | [16] | 16维观测状态，与action相同维度描述关节和夹爪位置 |
| observation.images.top | video | [480, 640, 3] | 顶部摄像头视频，分辨率480x640，3通道，AV1编码，30 FPS |
| observation.images.right_wrist | video | [480, 640, 3] | 右腕部摄像头视频，分辨率480x640，3通道，AV1编码，30 FPS |
| observation.images.left_wrist | video | [480, 640, 3] | 左腕部摄像头视频，分辨率480x640，3通道，AV1编码，30 FPS |
| timestamp | float32 | [1] | 时间戳 |
| frame_index | int64 | [1] | 帧索引 |
| episode_index | int64 | [1] | 片段索引 |
| index | int64 | [1] | 全局索引 |
| task_index | int64 | [1] | 任务索引 |
数据文件格式
- 数据路径:
data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet，每个 chunk 最多 1000 帧 - 视频路径:
videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4 - 配置文件:
meta/info.json包含完整的元数据定义
可视化
可通过以下链接交互式浏览数据集：
- 可视化入口：
https://huggingface.co/spaces/lerobot/visualize_dataset?path=access-robotics-team/fold_towel_merged_all
