遇见数据集

demo_src_mug_cleanup_task_D1_robot_Kinova3_gripper_Robotiq85Gripper

收藏
Hugging Face2026-06-25 更新2026-06-25 收录
官方服务:

资源简介:

该数据集是一个机器人学数据集,专门用于模仿学习和跨具身学习研究。它使用LeRobot工具创建,包含一个名为demo_src_mug_cleanup_task_D1的演示任务,涉及使用Kinova3机械臂和Robotiq85夹爪清理杯子。数据集包含100个episodes,总计33,659帧,帧率为20fps。数据以parquet文件格式存储,总数据大小约100MB,视频文件大小约200MB。数据集提供了丰富的观测和动作信息,包括:动作数据(7维浮点数组,表示机械臂的平移、旋转和夹爪控制);状态观测数据,如末端执行器位置和姿态(7维)、关节位置(7维)、夹爪状态(6维),以及组合观测(如末端执行器与夹爪的13维组合);图像观测数据,包括agentview和wrist视角的RGB视频(分辨率256x256,3通道,AV1编码)。此外,数据集还包含时间戳、帧索引、episode索引等元数据。所有数据仅用于训练分割(0:100)。该数据集适用于机器人控制、行为克隆和跨具身学习任务,遵循Apache 2.0许可证。

This dataset is a robotics dataset specifically designed for imitation learning and cross-embodiment learning research. It is created using the LeRobot tool and includes a demonstration task named demo_src_mug_cleanup_task_D1, which involves cleaning cups using a Kinova3 robotic arm and a Robotiq85 gripper. The dataset contains 100 episodes, totaling 33,659 frames at a frame rate of 20fps. Data is stored in parquet file format, with a total data size of approximately 100MB and video file size of about 200MB. The dataset provides rich observation and action information, including: action data (a 7-dimensional floating-point array representing the robotic arms translation, rotation, and gripper control); state observation data, such as end-effector position and pose (7-dimensional), joint positions (7-dimensional), gripper state (6-dimensional), and combined observations (e.g., a 13-dimensional combination of end-effector and gripper); image observation data, including RGB videos from agentview and wrist perspectives (resolution 256x256, 3 channels, AV1 encoding). Additionally, the dataset includes metadata such as timestamps, frame indices, and episode indices. All data is exclusively for training splits (0:100). This dataset is suitable for robot control, behavior cloning, and cross-embodiment learning tasks, and follows the Apache 2.0 license.

提供机构:
OliverHausdoerfer
创建时间:
2026-06-25
原始信息汇总

数据集概述

  • 数据集名称: demo_src_mug_cleanup_task_D1_robot_Kinova3_gripper_Robotiq85Gripper
  • 许可证: Apache-2.0
  • 任务类别: 机器人学(Robotics)
  • 标签: LeRobot, MimicGen, 跨具身(Cross-Embodiment)
  • 创建工具: 基于 LeRobot 创建

数据集规模

指标 数值
总片段数(Episodes) 100
总帧数(Frames) 33,659
总任务数(Tasks) 1
帧率(FPS) 20
数据文件大小 100 MB
视频文件大小 200 MB
数据块大小(Chunk Size) 1,000 帧/块
训练集划分 0:100(全部用于训练)

数据特征结构

数据集包含以下特征字段,每个字段的详细说明如下:

特征名 数据类型 形状 描述/分量名称
action float32 (7,) 动作指令:dx, dy, dz, drx, dry, drz, gripper(夹爪)
observation.state float32 (2,) 机器人状态:zero0, zero1
observation.eef float32 (7,) 末端执行器位姿:位置(x, y, z) + 四元数(x, y, z, w)
observation.gripper float32 (6,) 夹爪关节位置:gripper_qpos_0~5
observation.joint float32 (7,) 机器人关节位置:joint_pos_0~6
observation.eef_gripper float32 (13,) 末端执行器+夹爪组合(7+6维)
observation.joint_gripper float32 (13,) 关节+夹爪组合(7+6维)
observation.images.agentview 视频 (256,256,3) 代理视角RGB图像,格式AV1,20 FPS
observation.images.wrist 视频 (256,256,3) 腕部视角RGB图像,格式AV1,20 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

机器人信息

  • 机器人类型: 跨具身(Cross-Embodiment)
  • 任务: 杯子清理(Mug Cleanup Task)
  • 机器人型号: Kinova3
  • 夹爪型号: Robotiq85Gripper

数据存储路径

  • 数据文件: data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet
  • 视频文件: videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4
二维码
社区交流群
二维码
科研交流群
商业服务