遇见数据集

real01b_marker_d2_ours_sobol_r0

收藏
Hugging Face2026-07-24 更新2026-07-24 收录
官方服务:

资源简介:

这是一个Apache TsFile格式的转换数据集,源自LeRobot v3.0 Franka数据集,专门用于marker_d2任务。数据集包含时间序列数据,采样率为15帧/秒,仅提供训练分割,包含100个episodes,总计19,096帧。数据以TsFile表形式存储,包含94个列,涵盖时间戳、episode索引、任务索引、帧索引、动作向量(如关节和笛卡尔空间动作)、观察状态(如位置和速度)、标记任务几何数据、干预信号以及遥测数据(如Franka外部电机扭矩)。数据集未包含四个AV1相机流的视频文件,但保留了与原始视频的对齐信息。在转换过程中,所有源Parquet帧文件被合并为一个TsFile文件,并保留了原始元数据。该数据集适用于机器人学习任务,特别是基于时间序列的标记操作。

This is a converted dataset in Apache TsFile format, derived from the LeRobot v3.0 Franka dataset for the marker_d2 task. It contains time-series data at a sampling rate of 15 fps, with only a training split (100 episodes, totaling 19,096 frames). The data is stored in a TsFile table with 94 columns, covering timestamp, episode index, task index, frame index, action vectors (e.g., joint and Cartesian space actions), observation states (e.g., position and velocity), marker task geometric data, intervention signals, and telemetry data (e.g., Franka external motor torque). The dataset does not include video files from four AV1 camera streams but retains alignment information with the original videos. During conversion, all source Parquet frame files were merged into a single TsFile while preserving original metadata. It is suitable for robot learning tasks, particularly time-series-based marker manipulation.

提供机构:
zjt24
创建时间:
2026-07-17
二维码
社区交流群
二维码
科研交流群
商业服务