adapt1
收藏资源简介:
该机器人操作数据集使用LeRobot工具创建,采用Apache-2.0许可证。数据集包含一个完整的操作episode,总计208个时间帧(30 fps),对应1个任务和6个视频片段。数据以Parquet文件格式存储,视频以MP4格式存储。数据集记录了Droid机器人的操作过程,包含丰富的多模态观测和控制指令。具体特征包括:8维关节空间的动作和状态(关节1-7及夹爪)、10维的右末端执行器位姿(6D旋转、3D平移和夹爪状态)作为观测和动作目标、来自三个不同视角(左Azure Kinect、前Azure Kinect、腕部相机)的彩色图像(720x1280x3)和深度/对齐深度图像(720x1280x1),以及时间戳、帧索引、episode索引、数据索引和任务索引等元数据。该数据集适用于机器人模仿学习、视觉伺服控制、多模态策略学习等研究任务。
This robotic manipulation dataset is created using the LeRobot toolkit and licensed under the Apache-2.0 license. It contains one complete manipulation episode, totaling 208 timesteps at 30 fps, corresponding to 1 task and 6 video clips. The data is stored in Parquet file format, while the videos are saved in MP4 format. This dataset records the manipulation process of the Droid robot, and includes rich multimodal observations and control commands. Specific features include: 8-dimensional joint-space actions and states (joints 1-7 and the gripper); 10-dimensional right end-effector poses (6D rotation, 3D translation, and gripper state) serving as both observations and action targets; color images (720×1280×3) and depth/aligned depth images (720×1280×1) from three distinct viewpoints (left Azure Kinect, front Azure Kinect, and wrist camera); as well as metadata such as timestamps, frame index, episode index, data index, and task index. This dataset is suitable for research tasks including robotic imitation learning, visual servo control, and multimodal policy learning.
数据集概述:adapt1
该数据集是一个面向机器人领域的轻量级数据集,使用 LeRobot 框架创建。
- 许可协议:Apache-2.0
- 任务类别:机器人 (robotics)
- 标签:LeRobot
数据集规模与结构
- 总片段数 (Episodes):1
- 总帧数 (Frames):208
- 总任务数 (Tasks):1
- 总视频数 (Videos):6
- 帧率 (FPS):30
- 机器人类型:Droid
- 数据分割:仅包含训练集(
train),对应片段索引0:1 - 数据格式:Parquet 文件,存储路径为
data/*/*.parquet - 视频格式:MP4 文件
数据特征
数据集中包含多种特征,主要包括动作、状态观测和图像观测,具体如下:
| 特征名称 | 数据类型 | 形状 | 维度命名 | 说明 |
|---|---|---|---|---|
action |
float32 | [8] | joint_1 至 joint_7, gripper |
关节与夹爪的动作指令 |
observation.state |
float32 | [8] | joint_1 至 joint_7, gripper |
机器人关节与夹爪的状态反馈 |
observation.right_eef_pose |
float32 | [10] | rot_6d_0 至 rot_6d_5, trans_0 至 trans_2, gripper_articulation |
右端执行器的6D旋转、3D平移及夹爪关节位姿 |
action.right_eef_pose |
float32 | [10] | rot_6d_0 至 rot_6d_5, trans_0 至 trans_2, gripper_articulation |
右端执行器的6D旋转、3D平移及夹爪关节动作指令 |
observation.images.cam_azure_kinect_left.transformed_depth |
video | [720, 1280, 1] | height, width, channels |
左侧 Azure Kinect 变换后的深度图像(对齐到彩色图像) |
observation.images.cam_azure_kinect_left.color |
video | [720, 1280, 3] | height, width, channels |
左侧 Azure Kinect 原始彩色图像 |
observation.images.cam_azure_kinect_front.transformed_depth |
video | [720, 1280, 1] | height, width, channels |
前置 Azure Kinect 变换后的深度图像(对齐到彩色图像) |
observation.images.cam_azure_kinect_front.color |
video | [720, 1280, 3] | height, width, channels |
前置 Azure Kinect 原始彩色图像 |
observation.images.cam_wrist.color |
video | [720, 1280, 3] | height, width, channels |
腕部相机 RGB 彩色图像 |
observation.images.cam_wrist.depth |
video | [720, 1280, 1] | height, width, channels |
腕部相机深度图像 |
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 |




