rollout_stackblocks_iter1_correction_20260713_103916
收藏资源简介:
该数据集是一个机器人领域的数据集,使用LeRobot工具创建,专门用于训练和评估机器人控制模型,特别是涉及堆叠块(stackblocks)任务的校正数据。数据集包含机器人的动作控制数据、状态观测数据以及视觉观测数据。具体特征包括:动作(action)包含6个关节位置(肩部平移、肩部升降、肘部弯曲、腕部弯曲、腕部旋转、夹爪位置);状态观测(observation.state)同样包含这6个关节位置;视觉观测包括顶部摄像头图像(observation.images.top)和腕部摄像头图像(observation.images.wrist),均为RGB视频,分辨率480x640,帧率30fps。此外,数据集还包含干预标志(intervention)、下一帧成功标志(next.success)、时间戳(timestamp)、帧索引(frame_index)、episode索引(episode_index)、全局索引(index)和任务索引(task_index)。数据集总共有25个episodes,24489帧,数据以parquet文件格式存储,视频以mp4格式存储,总数据大小约100MB,视频大小约200MB。
This is a robotics dataset developed using the LeRobot tool, specifically dedicated to training and evaluating robot control models, particularly serving as calibration data for the block stacking (stackblocks) task. The dataset encompasses robot motion control data, state observation data, and visual observation data. Its specific features are as follows: the action field contains 6 joint positions, namely shoulder translation, shoulder lift, elbow flexion, wrist flexion, wrist rotation, and gripper position; the state observation (observation.state) also includes these 6 joint positions; the visual observations consist of top camera images (observation.images.top) and wrist camera images (observation.images.wrist), both being RGB videos with a resolution of 480x640 and a frame rate of 30 fps. Additionally, the dataset provides intervention flags, next success flags (next.success), timestamps, frame indexes, episode indexes (episode_index), global indexes (index), and task indexes (task_index). In total, the dataset contains 25 episodes and 24489 frames. The structured tabular data is stored in Parquet file format, while the video data is stored in MP4 format. The total size of the structured data is approximately 100 MB, and the total size of the video data is approximately 200 MB.
数据集概览
- 名称: rollout_stackblocks_iter1_correction_20260713_103916
- 许可协议: Apache-2.0
- 任务类别: 机器人学(robotics)
- 标签: LeRobot
数据集描述
该数据集使用 LeRobot 框架创建,包含机器人操作相关的演示数据,旨在用于模仿学习或机器人控制研究。
数据集结构
- 数据格式: Parquet 文件(
data/*/*.parquet),视频文件以 MP4 格式存储。 - 帧率 (FPS): 30
- 总集数 (Episodes): 25
- 总帧数 (Frames): 24,489
- 总任务数 (Tasks): 1
- 数据大小: 数据文件约 100 MB,视频文件约 200 MB
- 机器人类型:
so_follower - 数据分割: 所有数据用于训练(
train分割,包含 0-24 集)
特征字段
| 字段 | 类型 | 描述 |
|---|---|---|
action |
float32 (6维) | 机械臂关节位置动作(肩关节、肘关节、腕关节、夹爪等) |
observation.state |
float32 (6维) | 机械臂关节位置状态(与 action 维度对应) |
observation.images.top |
video (480x640x3) | 顶部摄像头图像,AV1 编码,30 FPS |
observation.images.wrist |
video (480x640x3) | 腕部摄像头图像,AV1 编码,30 FPS |
intervention |
bool | 是否有人工干预 |
next.success |
bool | 下一步是否成功 |
timestamp |
float32 | 时间戳 |
frame_index |
int64 | 帧索引 |
episode_index |
int64 | 集索引 |
index |
int64 | 全局索引 |
task_index |
int64 | 任务索引(仅一个任务) |
引用信息
引用信息尚待补充。



