task04_pour_liquid_from_tubes_to_beaker_gist
收藏资源简介:
该数据集是GIST AI Lab收集的11阶段实验室协议数据中的第4阶段,任务是将溶液从两个试管倒入最左边的烧杯。包含2040个episodes,共763573帧,帧率30 FPS,机器人类型为mobileai_robot,代码库版本v3.0。摄像头包括cam_high、cam_left_wrist、cam_right_wrist,分辨率480x640x3,编码格式av1。动作和观察状态为16维(移动基座2维+左臂7维+右臂7维)。同时提供了labels.csv文件,包含episode_id、task_scenario、success和data_issue列。所有2040个episodes均标记为成功,但其中651个存在data_issue(问题类型为mobile_base_not_captured,即移动基座状态未被正确记录),建议使用无data_issue且成功的1389个episodes进行训练。
This dataset is the 4th stage of the 11-stage laboratory protocol data collected by GIST AI Lab. The task is to pour solution from two test tubes into the leftmost beaker. It contains 2040 episodes with a total of 763,573 frames at 30 FPS. The robot type is mobileai_robot, codebase version v3.0. Cameras include cam_high, cam_left_wrist, and cam_right_wrist with resolution 480x640x3 and encoding format av1. The action and observation state are 16-dimensional (2 for mobile base, 7 for left arm, 7 for right arm). A labels.csv file is provided, containing columns episode_id, task_scenario, success, and data_issue. All 2040 episodes are marked as successful, but 651 of them have data_issue (type: mobile_base_not_captured, meaning the mobile base state was not correctly recorded). It is recommended to use the 1389 episodes that are both successful and without data_issue for training.
dataset详情:task04_pour_liquid_from_tubes_to_beaker_gist
基本信息
- 数据集名称:task04_pour_liquid_from_tubes_to_beaker_gist
- 来源机构:GIST AI Lab
- 许可证:Apache 2.0
- 任务类型:机器人(Robotics)
- 标签:LeRobot、mobile-ai
任务描述
- 任务说明:"Pour the solution from both tubes into the leftmost beaker"(将两个试管中的溶液倒入最左侧的烧杯中)
- 数据归属:本数据集为GIST AI Lab收集的11阶段实验室协议数据中的第4阶段
- 原始命名:
task4-pour-tubes-into-leftmost-beaker,经重命名后以task04_pour_liquid_from_tubes_to_beaker格式重新发布
数据规模
| 项目 | 数值 |
|---|---|
| Episodes(回合数) | 2,040 |
| Frames(总帧数) | 763,573 |
| FPS | 30 |
数据采集配置
- 机器人类型:mobileai_robot
- codebase版本:v3.0
- 摄像头(3个视角):
- cam_high
- cam_left_wrist
- cam_right_wrist
- 分辨率均为 480x640x3,编码格式为 av1
- action / observation.state维度:16D(Mobile AI——base 2D + 左臂7D + 右臂7D)
文件结构
- 数据文件格式:Parquet(
data/*/*.parquet) - 默认配置名称:default
labels.csv说明
该文件为原始发布中附带逐回合(episode)级标签,非LeRobot标准文件格式,独立存放于repo根目录。
| 类别 | episodes数量 |
|---|---|
| 总回合数 | 2,040 |
| success == 1 | 2,040 |
| 存在data_issue | 651 |
| 推荐用于训练(success == 1 且无 data_issue) | 1,389 |
data_issue说明
- 所有阶段的数据问题均为同一类型:
mobile_base_not_captured - 该问题表示:移动底座状态被记录为固定值,使base 2D数据失去意义
- 影响:仅使用手臂动作时无影响,但若将base数据纳入训练,则应筛选排除这些回合
训练筛选建议
在Python中可通过如下方式筛选推荐训练回合: import pandas as pd lab = pd.read_csv("labels.csv") keep = lab[(lab.success == 1) & (lab.data_issue.isna())] episodes = sorted(int(e.removeprefix("ep_")) for e in keep.episode_id)



