pusht-diffusion-policy-baseline-v1
收藏资源简介:
该数据集由LeRobot创建,旨在支持机器人学领域的相关研究,特别是视觉动作策略学习任务。数据集包含机器人交互数据,总规模为206个情节、25,650帧和206个视频,帧率为10 FPS。数据以训练集形式组织,涵盖单个任务。每个样本包含多个字段:观察图像(96x96x3的视频格式)、观察状态(2维浮点数,表示电机状态)、动作(2维浮点数,表示电机控制)、情节索引、帧索引、时间戳、下一状态奖励(浮点数)、下一状态完成标志(布尔值)、下一状态成功标志(布尔值)以及任务索引。数据格式为Parquet文件,适用于机器人控制、强化学习或模仿学习等应用场景。数据集基于MIT许可证发布,并与Diffusion Policy论文相关联。
This dataset is created by LeRobot and aims to support research in the field of robotics, particularly tasks related to visual action policy learning. It contains robot interaction data with a total scale of 206 episodes, 25,650 frames, and 206 videos at a frame rate of 10 FPS. The data is organized in a training set format, covering a single task. Each sample includes multiple fields: observation images (video format of 96x96x3), observation states (2-dimensional floating-point numbers representing motor states), actions (2-dimensional floating-point numbers representing motor control), episode index, frame index, timestamp, next state reward (floating-point number), next state done flag (boolean), next state success flag (boolean), and task index. The data format is Parquet files, suitable for applications such as robot control, reinforcement learning, or imitation learning. The dataset is released under the MIT license and is associated with the Diffusion Policy paper.
数据集概述
该数据集由韩国电子通信研究院(ETRI)的机器人AI平台开发团队创建,基于LeRobot框架生成,用于机器人技能学习任务。
基本信息
- 许可证:MIT(宽松开源许可证)
- 任务类别:机器人学(Robotics)
- 标签:LeRobot(Hugging Face机器人学习框架)
- 相关主页:Diffusion Policy 项目主页
- 相关论文:Diffusion Policy 论文
数据集规模
- 总片段数:206 个(episode)
- 总帧数:25,650 帧
- 总任务数:1 个
- 总视频数:206 个
- 帧率:10 FPS
- 数据集划分:全部206个片段用于训练(train: 0:206)
数据结构
数据集包含观察、动作和元数据等多种特征,数据以Parquet格式存储,视频以MP4格式存储。
特征详情表
| 特征名称 | 数据类型 | 形状 | 说明 |
|---|---|---|---|
| observation.image | 视频(AV1编码) | 96×96×3(高×宽×通道) | 机器人观测的RGB图像,10 FPS |
| observation.state | float32 | [2] | 机器人状态,包含2个电机(motor_0, motor_1) |
| action | float32 | [2] | 机器人动作,包含2个电机(motor_0, motor_1) |
| episode_index | int64 | [1] | 片段索引 |
| frame_index | int64 | [1] | 帧索引 |
| timestamp | float32 | [1] | 时间戳 |
| next.reward | float32 | [1] | 奖励值 |
| next.done | bool | [1] | 结束标志 |
| next.success | bool | [1] | 成功标志 |
| index | int64 | [1] | 全局索引 |
| task_index | int64 | [1] | 任务索引 |
引用
如需引用该数据集,推荐引用其基础工作《Diffusion Policy》论文:
bibtex @article{chi2024diffusionpolicy, author = {Cheng Chi and Zhenjia Xu and Siyuan Feng and Eric Cousineau and Yilun Du and Benjamin Burchfiel and Russ Tedrake and Shuran Song}, title ={Diffusion Policy: Visuomotor Policy Learning via Action Diffusion}, journal = {The International Journal of Robotics Research}, year = {2024}, }




