horse_picking
收藏资源简介:
该数据集由LeRobot创建,专注于机器人任务,具体涉及“horse_picking”(马匹拾取)场景。数据集包含50个episodes,总计25039帧,以parquet和mp4格式存储。数据特征包括:动作(6维浮点数组,表示机器人关节位置,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)、观察状态(与动作相同的6维关节位置)、来自手腕和头顶摄像头的图像观察(视频格式,分辨率480x640,3通道RGB,30fps),以及时间戳、帧索引、episode索引、任务索引等元数据。机器人类型为“so_follower”。数据集用于训练和评估机器人控制算法,支持强化学习或模仿学习任务。
This dataset is created by LeRobot and focuses on robotic tasks, specifically involving the horse_picking scenario. It contains 50 episodes, totaling 25039 frames, stored in parquet and mp4 formats. Data features include: actions (a 6-dimensional floating-point array representing robot joint positions, such as shoulder translation, shoulder elevation, elbow bend, wrist bend, wrist rotation, and gripper position), observation states (the same 6-dimensional joint positions as actions), image observations from wrist and overhead cameras (video format, resolution 480x640, 3-channel RGB, 30fps), as well as metadata such as timestamps, frame index, episode index, and task index. The robot type is so_follower. The dataset is used for training and evaluating robot control algorithms, supporting reinforcement learning or imitation learning tasks.
数据集概述
- 数据集名称:horse_picking
- 许可证:Apache-2.0
- 任务类别:机器人学(Robotics)
- 标签:LeRobot
数据集结构
- 代码库版本:v3.0
- 帧率:30 FPS
- 机器人类型:so_follower
特征(Features)
| 特征名称 | 数据类型 | 形状(尺寸) | 命名含义 |
|---|---|---|---|
action |
float32 | [6] | shoulder_pan.pos, shoulder_lift.pos, elbow_flex.pos, wrist_flex.pos, wrist_roll.pos, gripper.pos |
observation.state |
float32 | [6] | 同上 |
observation.images.wrist |
视频 | [480, 640, 3] | 高度×宽度×通道(RGB),编解码器:av1,像素格式:yuv420p,非深度图,30 FPS,无音频 |
observation.images.overhead |
视频 | [480, 640, 3] | 同上(俯视角) |
timestamp |
float32 | [1] | 时间戳 |
frame_index |
int64 | [1] | 帧索引 |
episode_index |
int64 | [1] | 回合索引 |
index |
int64 | [1] | 索引 |
task_index |
int64 | [1] | 任务索引 |
数据集规模
- 总回合数:50
- 总帧数:25,039
- 总任务数:1
- 分块大小:1,000
- 数据文件大小(Parquet):约100 MB
- 视频文件大小:约200 MB
数据路径
- 数据文件:
data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet - 视频文件:
videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4
数据集划分
- 训练集:回合 0 至 49(共50回合)



