record-test-3
收藏资源简介:
该数据集由LeRobot项目创建,属于机器人学领域,采用Apache-2.0许可证。它包含机器人控制相关的演示数据,针对so_follower机器人类型,规模为20个完整episodes,共8978帧数据,涵盖1个任务。数据以分块形式存储,包括结构化数据文件(parquet格式,约100MB)和视频文件(mp4格式,约200MB,帧率30fps)。数据字段包括:动作空间(6维浮点向量,表示机器人六个关节的位置)、观测状态(6维浮点向量,与动作空间对应)、前视相机图像观测(480x640分辨率RGB视频),以及时间戳、帧索引、episode索引、全局索引和任务索引等元数据。数据集仅提供训练集划分(全部20个episodes),适用于机器人模仿学习、行为克隆、离线强化学习等任务。
This dataset is created by the LeRobot project, belonging to the field of robotics and licensed under Apache-2.0. It contains demonstration data related to robot control, specifically for the so_follower robot type. The dataset includes 20 complete episodes, totaling 8978 frames of data, covering 1 task. It is stored in chunked form, comprising structured data files (parquet format, approximately 100MB) and video files (mp4 format, approximately 200MB, with a frame rate of 30fps). Data fields include: action space (a 6-dimensional floating-point vector representing the positions of six robot joints: shoulder translation, shoulder lift, elbow bend, wrist bend, wrist rotation, and gripper position); observation state (a 6-dimensional floating-point vector corresponding to the joint positions in the action space); front camera image observation (480x640 resolution RGB video); as well as metadata such as timestamps, frame index, episode index, global index, and task index. The dataset only provides a training set split (all 20 episodes) and is suitable for tasks like robot imitation learning, behavior cloning, and offline reinforcement learning.
数据集概述
- 数据集名称: record-test-3
- 许可证: Apache-2.0
- 任务类别: 机器人 (robotics)
- 标签: LeRobot
数据集详情
- 机器人类型: so_follower
- 总片段数 (episodes): 20
- 总帧数 (frames): 8978
- 总任务数 (tasks): 1
- 帧率 (fps): 30
- 数据文件大小: 100 MB
- 视频文件大小: 200 MB
- 数据集划分: 训练集 (train) 包含全部20个片段 (0-20索引)
数据结构
数据集包含以下特征字段:
-
action (动作):
- 数据类型: float32
- 维度: 6
- 包含关节位置: shoulder_pan.pos, shoulder_lift.pos, elbow_flex.pos, wrist_flex.pos, wrist_roll.pos, gripper.pos
-
observation.state (观测状态):
- 数据类型: float32
- 维度: 6
- 包含关节位置: shoulder_pan.pos, shoulder_lift.pos, elbow_flex.pos, wrist_flex.pos, wrist_roll.pos, gripper.pos
-
observation.images.front (前视图像):
- 数据类型: 视频 (video)
- 分辨率: 480x640,3通道
- 编码格式: av1,yuv420p
- 帧率: 30 fps
- 非深度图,无音频
-
timestamp (时间戳): float32,标量
-
frame_index (帧索引): int64,标量
-
episode_index (片段索引): int64,标量
-
index (索引): int64,标量
-
task_index (任务索引): int64,标量
数据存储格式
- 数据文件路径:
data/*/*.parquet(按块和文件索引组织) - 视频文件路径:
videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4
其他信息
- 代码库版本: v3.0
- 数据块大小: 1000
- 引用信息: 暂无格式引用内容




