five

cube_bowl_2markers_v3_clean

收藏
Hugging Face2026-06-07 更新2026-06-07 收录
下载链接:
https://huggingface.co/datasets/pdhinaka/cube_bowl_2markers_v3_clean
下载链接
链接失效反馈
官方服务:
资源简介:
该数据集使用LeRobot创建,包含机器人操作任务的数据,具体涉及立方体和碗的交互。数据集包括动作和观察特征:动作特征为6个关节位置(肩部平移、肩部提升、肘部弯曲、腕部弯曲、腕部旋转、夹爪位置);观察特征包括状态观察(6个关节位置)和图像观察(顶部和手腕视角的RGB视频,分辨率480x640,帧率30fps)。此外,数据集包含时间戳、帧索引、episode索引、任务索引等元数据。数据集共有50个episodes,总帧数13903帧,涉及2个任务,数据以parquet格式存储,视频以mp4格式存储,机器人类型为so_follower。数据集用于训练和评估机器人控制模型,支持模仿学习或强化学习任务。

This dataset was created using LeRobot and contains data for robot manipulation tasks, specifically involving interactions with cubes and bowls. The dataset includes action and observation features: action features consist of 6 joint positions (shoulder translation, shoulder elevation, elbow bend, wrist bend, wrist rotation, gripper position); observation features include state observations (6 joint positions) and image observations (RGB videos from top and wrist perspectives, with a resolution of 480x640 and a frame rate of 30fps). Additionally, the dataset contains metadata such as timestamps, frame indices, episode indices, and task indices. There are 50 episodes in total, with a total of 13903 frames, covering 2 tasks. The data is stored in parquet format, videos are in mp4 format, and the robot type is so_follower. The dataset is intended for training and evaluating robot control models, supporting imitation learning or reinforcement learning tasks.
提供机构:
pdhinaka
创建时间:
2026-06-07
原始信息汇总
  • 数据集名称:cube_bowl_2markers_v3_clean
  • 许可证:Apache-2.0
  • 任务类别:机器人技术(robotics)
  • 标签:LeRobot
  • 创建方式:使用 LeRobot 框架创建

数据集结构

  • 代码库版本:v3.0
  • 帧率(FPS):30
  • 机器人类型:so_follower
  • 总片段数(episodes):50
  • 总帧数:13,903
  • 总任务数:2
  • 数据分块大小:1,000
  • 数据文件大小:约100 MB
  • 视频文件大小:约200 MB
  • 数据路径格式data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet
  • 视频路径格式videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4
  • 数据集划分:训练集(train)包含全部50个片段(索引0:50)

特征描述

特征名 数据类型 形状 说明
action float32 [6] 机器人动作指令:肩部旋转(shoulder_pan.pos)、肩部升降(shoulder_lift.pos)、肘部弯曲(elbow_flex.pos)、腕部弯曲(wrist_flex.pos)、腕部旋转(wrist_roll.pos)、夹爪位置(gripper.pos)
observation.state float32 [6] 机器人观测状态,维度名称与action相同
observation.images.top 视频(video) [480, 640, 3] 顶部摄像头图像,分辨率480×640,3通道,编码为av1,像素格式yuv420p,帧率30 FPS
observation.images.wrist 视频(video) [480, 640, 3] 腕部摄像头图像,分辨率480×640,3通道,编码为av1,像素格式yuv420p,帧率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
二维码
社区交流群
二维码
科研交流群
商业服务