bridge_orig
收藏资源简介:
该数据集是一个大规模机器人示教数据集,使用LeRobot框架创建,专门针对机器人学习任务。数据采集自WidowX机器人,遵循RLDS(机器人学习数据集标准)格式。数据集核心内容为机器人执行任务的交互记录,包含丰富的多模态信息。数据规模庞大,总计包含53,191条轨迹(episodes),约199.9万帧数据,覆盖19,973个独立任务。数据以分块Parquet文件形式组织,并附带对应的视频文件。主要特征包括:1) 观测部分:提供四个同步的RGB摄像头图像观测(image_0, image_1, image_2, image_3),每帧图像分辨率为256x256,编码格式为AV1,帧率为5fps。2) 状态部分:记录机器人末端执行器的三维位置(eef_xyz)、欧拉角姿态(eef_rpy)、6D旋转表示(eef_rot6d)以及夹爪开合状态(gripper_state)。同时提供了一个整合的状态向量(observation.state)。3) 动作部分:提供了多种表示形式的动作指令,包括末端执行器位置的差分指令(diff_eef_xyz)、姿态差分指令(diff_eef_rpy)、在世界坐标系下的目标位置(world_eef_xyz)和6D旋转(world_eef_rot6d)、在夹爪坐标系下的目标位置(gripper_eef_xyz)和6D旋转(gripper_eef_rot6d)、夹爪目标状态(gripper_state),以及一个整合的动作向量(action)。4) 索引信息:包含时间戳(timestamp)、帧索引(frame_index)、轨迹索引(episode_index)、数据索引(index)和任务索引(task_index)。该数据集适用于机器人模仿学习、离线强化学习、视觉运动策略学习、多视角视觉表示学习等研究任务。数据集采用Apache-2.0许可证。
This is a large-scale robot demonstration dataset created using the LeRobot framework, specifically designed for robot learning tasks. The data is collected from the WidowX robot and adheres to the RLDS (Robot Learning Dataset Standard) format. The core content of the dataset consists of interaction records of robots performing tasks, containing rich multimodal information. The dataset has a massive scale, with a total of 53,191 episodes, approximately 1.999 million frames of data, covering 19,973 independent tasks. The data is organized into chunked Parquet files, with corresponding video files provided alongside. The main features are as follows: 1) Observation section: Four synchronized RGB camera image observations (image_0, image_1, image_2, image_3) are provided. Each frame has a resolution of 256×256, encoded in AV1 format, with a frame rate of 5 fps. 2) State section: Records the 3D position (eef_xyz), Euler angle pose (eef_rpy), 6D rotation representation (eef_rot6d) of the robot's end-effector, as well as the gripper state (gripper_state). An integrated state vector (observation.state) is also provided. 3) Action section: Provides action instructions in multiple representation forms, including differential end-effector position commands (diff_eef_xyz), differential pose commands (diff_eef_rpy), target position in the world coordinate system (world_eef_xyz) and 6D rotation (world_eef_rot6d), target position in the gripper coordinate system (gripper_eef_xyz) and 6D rotation (gripper_eef_rot6d), gripper state (gripper_state), as well as an integrated action vector (action). 4) Index information: Includes timestamp, frame_index, episode_index, data index (index), and task_index. This dataset is applicable to research tasks such as robot imitation learning, offline reinforcement learning, visuomotor policy learning, and multi-view visual representation learning. The dataset is licensed under the Apache-2.0 license.
数据集概述
- 数据集名称: bridge_orig
- 许可证: Apache-2.0
- 任务类别: 机器人学 (Robotics)
- 标签: LeRobot, bridge_orig, rlds, openx, widowx
数据集规模
- 总片段数: 53,191
- 总帧数: 1,999,362
- 总任务数: 19,973
- 数据文件大小: 100 MB
- 视频文件大小: 200 MB
- 帧率: 5 FPS
数据集划分
- 训练集: 片段 0 至 53,190(全部数据)
机器人类型
- 机器人类型: WidowX
数据特征
该数据集包含以下特征:
-
观测图像:
observation.images.image_0: 视频,尺寸 256x256x3,RGB,编码 AV1,帧率 5 FPSobservation.images.image_1: 视频,尺寸 256x256x3,RGB,编码 AV1,帧率 5 FPSobservation.images.image_2: 视频,尺寸 256x256x3,RGB,编码 AV1,帧率 5 FPSobservation.images.image_3: 视频,尺寸 256x256x3,RGB,编码 AV1,帧率 5 FPS
-
机器人状态:
state.eef_xyz: 末端执行器位置 (x, y, z),float32state.eef_rpy: 末端执行器姿态 (roll, pitch, yaw),float32state.eef_rot6d: 末端执行器旋转 6D 表示,float32state.gripper_state: 夹爪状态,float32observation.state: 完整状态 (x, y, z, rot1-6, gripper),float32
-
动作空间 (均为 float32):
action.diff_eef_xyz: 末端执行器位置差 (x, y, z)action.diff_eef_rpy: 末端执行器姿态差 (roll, pitch, yaw)action.world_eef_xyz: 世界坐标系下的末端执行器位置action.world_eef_rot6d: 世界坐标系下的末端执行器旋转 6D 表示action.gripper_eef_xyz: 夹爪坐标系下的末端执行器位置action.gripper_eef_rot6d: 夹爪坐标系下的末端执行器旋转 6D 表示action.gripper_state: 夹爪动作action: 完整动作向量 (x, y, z, rot1-6, gripper)
-
其他特征:
timestamp: 时间戳,float32frame_index: 帧索引,int64episode_index: 片段索引,int64index: 索引,int64task_index: 任务索引,int64
数据格式
- 数据文件路径:
data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet - 视频文件路径:
videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4
创建信息
- 该数据集使用 LeRobot 创建。




