autoresearch-vibepi-000250-cucumber-grab
收藏资源简介：
该数据集是一个用于机器人学习的多模态数据集，由LeRobot工具创建。数据集包含机器人状态观测（7个浮点数表示的关节位置，包括肩部平移、肩部抬升、肘部弯曲、腕部弯曲、腕部旋转、夹爪位置和倾斜位置）、来自三个视角（基础、顶部和腕部）的视频图像观测（分辨率480x640，3通道，20fps，AV1编码），以及相应的动作数据（7个浮点数表示的关节位置控制）。此外，数据集还包括时间戳、帧索引、episode索引、任务索引等元数据。数据集总共有1个episode、163帧和1个任务，数据以parquet格式存储，视频以mp4格式存储，适用于训练和评估机器人控制模型。机器人类型为vibeboard_follower_tilt，许可证为Apache-2.0。
This multimodal dataset for robotics learning was developed using the LeRobot toolkit. It contains robot state observations represented by 7 floating-point numbers for joint positions: shoulder translation, shoulder elevation, elbow flexion, wrist flexion, wrist rotation, gripper position, and tilt position. It also includes video image observations from three viewpoints (base, top, and wrist), with a resolution of 480x640, 3 color channels, 20 frames per second, and encoded in AV1 format, alongside corresponding action data consisting of 7 floating-point numbers for joint position control. Additionally, the dataset provides metadata such as timestamps, frame indices, episode indices, and task indices. In total, the dataset comprises 1 episode, 163 frames, and 1 task. The tabular data is stored in Parquet format while the video files are stored in MP4 format, and it is applicable for training and evaluating robotic control models. The robot type is vibeboard_follower_tilt, and the license is Apache-2.0.
数据集概述
- 数据集名称: autoresearch-vibepi-000250-cucumber-grab
- 数据集地址: https://huggingface.co/datasets/VibeCuisine/autoresearch-vibepi-000250-cucumber-grab
- 许可证: Apache-2.0
- 任务类别: 机器人学 (Robotics)
- 创建工具: LeRobot (https://github.com/huggingface/lerobot)
数据集结构
- 总集数 (Total Episodes): 1
- 总帧数 (Total Frames): 163
- 总任务数 (Total Tasks): 1
- 帧率 (FPS): 20
- 数据文件大小: 100 MB
- 视频文件大小: 200 MB
- 机器人类型: vibeboard_follower_tilt
特征 (Features)
观测状态 (observation.state)
- 数据类型: float32
- 形状: [7]
- 名称: shoulder_pan.pos, shoulder_lift.pos, elbow_flex.pos, wrist_flex.pos, wrist_roll.pos, gripper.pos, tilt.pos
观测图像 (observation.images)
包含三个摄像头视角，每个视角的视频规格如下：
- 分辨率: 480x640 像素 (高度 x 宽度)
- 通道数: 3 (RGB)
- 视频编码: AV1
- 像素格式: yuv420p
- 帧率: 20 FPS
- 是否为深度图: 否
摄像头视角列表：
- base: 基础摄像头
- top: 顶部摄像头
- wrist: 腕部摄像头
动作 (action)
- 数据类型: float32
- 形状: [7]
- 名称: shoulder_pan.pos, shoulder_lift.pos, elbow_flex.pos, wrist_flex.pos, wrist_roll.pos, gripper.pos, tilt.pos
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]
数据划分 (Splits)
- 训练集 (train): 0:1 (全部数据用于训练)
数据存储路径
- 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
