遇见数据集

rollout_2026-06-10_shakeitup_aluboxvf_0_10kHz_check10k_20260610_161535

收藏
Hugging Face2026-06-10 更新2026-06-10 收录
官方服务:

资源简介:

该数据集是一个机器人学数据集,使用LeRobot工具创建,专注于机器人动作和观察数据的收集。数据集包含10个episodes,总计4543帧,数据以Parquet文件格式存储,视频文件以MP4格式存储。数据集的结构包括:动作数据(action),为6维浮点数组,对应机器人关节位置(如肩部平移、肩部提升、肘部弯曲、手腕弯曲、手腕旋转和夹爪位置);观察数据(observation),包括状态(state,同样为6维关节位置)、来自顶部摄像头(top)和手腕摄像头(wrist)的图像(分辨率480x640,3通道,视频编码为av1,帧率30fps),以及10kHz触觉频谱图图像(observation.images.10kHz_tactile_spectrogram,分辨率224x224,3通道)。此外,数据集还包含时间戳、帧索引、episode索引、索引和任务索引等元数据。机器人类型为so101_follower_dragontactile,数据集适用于训练和评估机器人控制模型,支持机器人学任务如动作预测和环境感知。数据分块存储,每个块大小为1000帧,总数据文件大小为100MB,视频文件大小为200MB。

This is a robotics dataset created using the LeRobot tool, which focuses on collecting robotic action and observation data. It contains 10 episodes with a total of 4543 frames. The dataset's data is stored in Parquet file format, while video files are saved in MP4 format. The dataset structure includes action data (`action`), a 6-dimensional floating-point array corresponding to robot joint positions including shoulder translation, shoulder elevation, elbow flexion, wrist flexion, wrist rotation and gripper position; observation data (`observation`), which includes the state (`state`, also a 6-dimensional joint position array), images from the top camera (`top`) and wrist camera (`wrist`) with a resolution of 480x640, 3 channels, encoded with AV1 video codec at 30fps, as well as 10kHz tactile spectrogram images at `observation.images.10kHz_tactile_spectrogram` with a resolution of 224x224 and 3 channels. In addition, the dataset also includes metadata such as timestamps, frame indices, episode indices, indexes and task indices. The robot type is so101_follower_dragontactile, and the dataset is applicable to training and evaluating robot control models, supporting robotics tasks such as action prediction and environment perception. The data is stored in chunks, with each chunk containing 1000 frames, and the total size of the data files is 100 MB while the total size of the video files is 200 MB.

提供机构:
jogarulfop
创建时间:
2026-06-10
原始信息汇总

数据集概述

该数据集是一个用于机器人学的演示数据集,使用 LeRobot 框架创建,遵循 Apache-2.0 许可。

数据集结构

  • 总片段数 (total_episodes): 10
  • 总帧数 (total_frames): 4543
  • 总任务数 (total_tasks): 1
  • 帧率 (fps): 30
  • 代码库版本 (codebase_version): v3.0
  • 机器人类型 (robot_type): so101_follower_dragontactile
  • 数据划分 (splits): 所有 10 个片段均用于训练 (train: "0:10")
  • 数据文件大小: 约 100 MB (parquet 格式)
  • 视频文件大小: 约 200 MB (mp4 格式)

数据特征

每条数据包含以下特征:

特征名称 数据类型 形状 描述
action float32 [6] 机器人动作,包含 6 个关节的位置信息。
observation.state float32 [6] 机器人观察到的状态,与动作空间一致。
observation.images.top video [480, 640, 3] 顶部视角摄像头,分辨率 480x640,AV1 编码。
observation.images.wrist video [480, 640, 3] 腕部视角摄像头,分辨率 480x640,AV1 编码。
observation.images.10kHz_tactile_spectrogram video [224, 224, 3] 10kHz 触觉频谱图,分辨率 224x224,AV1 编码。
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。

动作/状态空间

actionobservation.state 均为 6 维向量,分别对应机器人关节的位置:

  1. shoulder_pan.pos
  2. shoulder_lift.pos
  3. elbow_flex.pos
  4. wrist_flex.pos
  5. wrist_roll.pos
  6. gripper.pos
二维码
社区交流群
二维码
科研交流群
商业服务