full_featured
收藏资源简介：
该数据集由LeRobot平台创建，是一个用于机器人技术研究的开源数据集。数据集记录了机器人执行任务过程中的多模态交互数据，包含68个完整任务片段（episodes），共计75,257帧数据，覆盖26种不同的任务类型。数据以Parquet文件格式存储，总数据文件大小为100MB，视频文件大小为200MB。数据集包含丰富的特征字段：动作空间（action）包含5维浮点向量，表示机器人在x、y、z方向的速度以及手腕和手指的运动速度；观测状态（observation.state）是一个43维浮点向量，详细记录了机器人的状态信息，包括夹爪位置与旋转、传感器读数（激光测距、手指压力、张力等）、目标物体相对位置（如障碍物、玩具箱、垃圾桶等）以及视觉定位坐标；图像观测（observation.images）提供三个摄像头的视频流：一个夹爪摄像头（分辨率384×384）和两个锚定摄像头（分辨率均为544×960），所有视频均为30fps的RGB格式。此外，数据集还包含时间戳、帧索引、episode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、强化学习、多模态感知与控制等研究任务，仅提供训练集划分，采用Apache 2.0开源许可证。
This open-source dataset was created by the LeRobot platform and tailored for robotics research. It records multimodal interaction data during robot task execution, consisting of 68 complete task episodes totaling 75,257 frames, covering 26 distinct task types. The data is stored in Parquet file format, with a total data file size of 100 MB and a total video file size of 200 MB. The dataset includes rich feature fields: the action space (action) is a 5-dimensional floating-point vector representing the robot's velocities in the x, y, and z directions, as well as the movement speeds of the wrist and fingers; the observation state (observation.state) is a 43-dimensional floating-point vector that comprehensively records the robot's status information, including gripper position and rotation, sensor readings such as laser rangefinder data, finger pressure, tension, etc., relative positions of target objects (e.g., obstacles, toy boxes, trash cans, etc.), and visual positioning coordinates. The image observations (observation.images) provide video streams from three cameras: one wrist-mounted gripper camera with a resolution of 384×384 and two fixed anchor cameras both with a resolution of 544×960, with all videos being in 30fps RGB format. In addition, the dataset also includes metadata such as timestamps, frame indices, episode indices, and task indices. This dataset is applicable to research tasks such as robot imitation learning, reinforcement learning, multimodal perception and control, and only provides a training set split, and is licensed under the Apache 2.0 open-source license.
- 许可证: Apache-2.0
- 任务类别: 机器人学 (robotics)
- 标签: LeRobot
- 数据集来源: 使用 LeRobot 创建
数据集描述
- 机器人类型: stringman
- 总片段数 (episodes): 68
- 总帧数 (frames): 75,257
- 总任务数 (tasks): 26
- 块大小 (chunks_size): 1000
- 数据文件大小: 100 MB
- 视频文件大小: 200 MB
- 帧率 (FPS): 30
- 训练/测试划分: 全部 68 个片段用于训练 (splits: {"train": "0:68"})
数据特征 (Features)
动作 (action)
- 数据类型: float32
- 形状: [5]
- 字段名称:
vel_x,vel_y,vel_z,wrist_speed,finger_speed
观测状态 (observation.state)
- 数据类型: float32
- 形状: [43]
- 包含:
- 速度:
vel_x,vel_y,vel_z - 腕部/手指:
wrist_speed,finger_speed,gripper_pos_x,gripper_pos_y,gripper_pos_z,gripper_rot_0..gripper_rot_5,finger_angle,finger_pressure,wrist_angle - 传感器:
laser_rangefinder,target_force - 物体方位:
hamper_bearing,hamper_distance,toybox_bearing,toybox_distance,trashcan_bearing,trashcan_distance,gamepad_bearing,gamepad_distance,parking_location_bearing,parking_location_distance - 控制状态:
swing_cancellation_on,tension_0..tension_3 - 位置坐标:
gantry_position_x,gantry_position_y,gantry_position_z,visual_pos_x,visual_pos_y,visual_pos_z,hang_pos_x,hang_pos_y,hang_pos_z
- 速度:
观测图像 (observation.images)
- 夹爪相机 (gripper_camera):
- 数据类型: video
- 分辨率: 384 x 384
- 编码: AV1, 30 FPS, 3 通道
- 锚点相机1 (anchor_camera_1):
- 数据类型: video
- 分辨率: 544 x 960
- 编码: AV1, 30 FPS, 3 通道
- 锚点相机2 (anchor_camera_2):
- 数据类型: video
- 分辨率: 544 x 960
- 编码: AV1, 30 FPS, 3 通道
其他特征
- timestamp: float32, 形状 [1]
- frame_index: int64, 形状 [1]
- episode_index: int64, 形状 [1]
- index: int64, 形状 [1]
- task_index: int64, 形状 [1]
