so101_cube_pick_place_2cam_ws_0625
收藏资源简介：
so101_cube_pick_place_0514 是一个用于机器人学习的示范数据集，专门针对方块拾取与放置任务。该数据集使用LeRobot框架创建，包含机器人执行单一任务（方块拾取放置）的交互数据。数据规模方面，数据集共包含101个完整的情节（episodes），总计184,117帧数据。数据文件总大小约为100MB，关联的视频文件总大小约为200MB。数据以分块形式组织，每个数据块包含1000帧，存储为Parquet格式；视频则以MP4格式存储，帧率为30fps。数据划分上，所有101个情节均用于训练。数据内容包含多模态信息：动作空间为6维浮点数组，对应机器人肩部平移、肩部抬升、肘部弯曲、腕部弯曲、腕部旋转和夹爪的位置控制。观测空间包含两部分：一是与动作空间维度相同的机器人关节状态（6维浮点数）；二是两个相机视角的图像观测——腕部相机和侧面相机，均提供480x640分辨率、3通道（RGB）的视频流。此外，数据还包含时间戳、帧索引、情节索引、全局索引和任务索引等元数据字段。该数据集适用于机器人模仿学习、强化学习、行为克隆以及视觉-动作策略学习等任务，尤其专注于机械臂操作场景下的拾取放置技能学习。
so101_cube_pick_place_0514 is a demonstration dataset for robot learning, specifically designed for cube pick-and-place tasks. It is created using the LeRobot framework and contains interaction data of robots performing a single task (cube pick-and-place). In terms of data scale, the dataset includes 101 complete episodes, totaling 184,117 frames. The total size of the data files is approximately 100MB, with associated video files totaling about 200MB. The data is organized in chunks, with each chunk containing 1,000 frames stored in Parquet format; videos are stored in MP4 format with a frame rate of 30fps. For data partitioning, all 101 episodes are used for training. The data content includes multimodal information: the action space is a 6-dimensional floating-point array corresponding to robot shoulder translation, shoulder lift, elbow bend, wrist bend, wrist rotation, and gripper position control. The observation space consists of two parts: one is the robot joint state (6-dimensional floating-point numbers) with the same dimensions as the action space; the other is image observations from two camera perspectives—a wrist camera and a side camera, both providing 480x640 resolution, 3-channel (RGB) video streams. Additionally, the data includes metadata fields such as timestamps, frame indices, episode indices, global indices, and task indices. This dataset is suitable for tasks such as robot imitation learning, reinforcement learning, behavior cloning, and vision-action policy learning, with a particular focus on pick-and-place skill learning in robotic arm manipulation scenarios.
数据集概述
- 数据集名称：so101_cube_pick_place_2cam_ws_0625
- 任务类别：机器人技术 (robotics)
- 许可协议：Apache-2.0
- 创建工具：LeRobot
- 总片段数 (episodes)：101
- 总帧数 (frames)：184,117
- 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
- 训练集划分：所有片段（0:101）均用于训练
数据结构
数据集以 Parquet 文件存储数据，以 MP4 文件存储视频，组织方式如下：
- 数据路径：
data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet（每组最大 1000 帧） - 视频路径：
videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4
特征 (Features)
| 特征名称 | 数据类型 | 形状 | 说明 |
|---|---|---|---|
action |
float32 | (6,) | 动作指令，包含6个关节位置：shoulder_pan, shoulder_lift, elbow_flex, wrist_flex, wrist_roll, gripper |
observation.state |
float32 | (6,) | 观察状态，与动作维度一致，反映机器人当前关节位置 |
observation.images.wrist |
video | (480, 640, 3) | 腕部摄像头视频，分辨率480×640，AV1编码，30 fps |
observation.images.side |
video | (480, 640, 3) | 侧方摄像头视频，分辨率480×640，AV1编码，30 fps |
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 |
机器人类型
- 机器人型号：so_follower
- 动作/状态空间维度：6（包含肩部、肘部、腕部及夹爪的关节位置）
