cmu_stretch
收藏资源简介:
该数据集是通过LeRobot创建的,主要用于机器人学习任务。数据集包含135个episodes,25016帧数据,5个任务。数据包括观察图像(128x128x3)、语言指令、机器人状态(4个电机状态)、动作(8个电机状态)、时间戳、episode索引、帧索引、奖励、完成标志等。数据格式为parquet,视频格式为mp4,帧率为5fps。数据集适用于机器人视觉和动作学习研究。
This dataset was developed via LeRobot, primarily targeting robotic learning tasks. It contains 135 episodes, 25016 frames of data, and covers 5 distinct tasks. The included data modalities are observed images (128×128×3), language instructions, robot states (4 motor states), actions (8 motor states), timestamps, episode indices, frame indices, rewards, completion flags, and other relevant information. The data is stored in Parquet format, while the supporting videos are in MP4 format with a frame rate of 5 fps. This dataset is suitable for research in robotic vision and action learning.
数据集概述
基本信息
- 名称: cmu_stretch
- 主页: https://robo-affordances.github.io/
- 论文: https://arxiv.org/abs/2304.08488
- 许可证: MIT
- 任务类别: 机器人学
- 标签: LeRobot
数据集结构
- 总片段数: 135
- 总帧数: 25016
- 总任务数: 5
- 总视频数: 135
- 总块数: 1
- 块大小: 1000
- 帧率: 5 fps
- 分割: 训练集 (0:135)
数据文件路径
- 数据路径: data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet
- 视频路径: videos/chunk-{episode_chunk:03d}/{video_key}/episode_{episode_index:06d}.mp4
特征
-
observation.images.image
- 类型: 视频
- 形状: [128, 128, 3]
- 视频信息:
- 帧率: 5.0
- 编解码器: av1
- 像素格式: yuv420p
- 深度图: 否
- 音频: 无
-
language_instruction
- 类型: 字符串
- 形状: [1]
-
observation.state
- 类型: float32
- 形状: [4]
- 名称: motor_0, motor_1, motor_2, motor_3
-
action
- 类型: float32
- 形状: [8]
- 名称: motor_0, motor_1, motor_2, motor_3, motor_4, motor_5, motor_6, motor_7
-
timestamp
- 类型: float32
- 形状: [1]
-
episode_index
- 类型: int64
- 形状: [1]
-
frame_index
- 类型: int64
- 形状: [1]
-
next.reward
- 类型: float32
- 形状: [1]
-
next.done
- 类型: bool
- 形状: [1]
-
index
- 类型: int64
- 形状: [1]
-
task_index
- 类型: int64
- 形状: [1]
引用
BibTeX
bibtex @inproceedings{bahl2023affordances, title={Affordances from Human Videos as a Versatile Representation for Robotics}, author={Bahl, Shikhar and Mendonca, Russell and Chen, Lili and Jain, Unnat and Pathak, Deepak}, booktitle={CVPR}, year={2023} }
@article{mendonca2023structured, title={Structured World Models from Human Videos}, author={Mendonca, Russell and Bahl, Shikhar and Pathak, Deepak}, journal={CoRL}, year={2023} }




