UR5e_sim_pick_place_can_100_absolute_EEF_actions
收藏资源简介:
该数据集是使用LeRobot工具包创建的机器人仿真数据集。数据集聚焦于UR5e机器人在robosuite仿真环境中的拾放任务,数据通过脚本化方式生成。数据集包含100个完整的任务回合,总计13495个数据帧,帧率为20fps。每个数据样本包含机器人的动作指令(一个4维浮点向量)、机器人的状态观测(一个13维浮点向量)、以及来自两个固定摄像头(camera1和camera2)的视觉观测(视频格式,分辨率为256x256的RGB图像)。此外,每个样本还包含时间戳、帧索引、回合索引和任务索引等元数据。数据集以Parquet文件格式组织,并包含对应的视频文件。所有数据被划分为训练集。该数据集适用于机器人学习任务,特别是用于训练和评估基于视觉的机器人操作策略,如模仿学习或强化学习。
This dataset is a robot simulation dataset created using the LeRobot toolkit. It focuses on the pick-and-place task of a UR5e robot in the robosuite simulation environment, with data generated through scripting. The dataset contains 100 complete task episodes, totaling 13,495 data frames at a frame rate of 20 fps. Each data sample includes the robots action command (a 4-dimensional floating-point vector), the robots state observation (a 13-dimensional floating-point vector), and visual observations from two fixed cameras (camera1 and camera2) in video format (256x256 RGB images). Additionally, each sample contains metadata such as timestamps, frame indices, episode indices, and task indices. The dataset is organized in Parquet file format and includes corresponding video files. All data is split into a training set. This dataset is suitable for robot learning tasks, particularly for training and evaluating vision-based robot manipulation policies such as imitation learning or reinforcement learning.
- 数据集名称:UR5e_sim_pick_place_can_100_absolute_EEF_actions
- 许可证:Apache-2.0
- 任务类别:机器人学(Robotics)
- 标签:LeRobot、UR5e、Robosuite、Pick-Place、Scripted
- 数据集规模:包含100个任务片段(episodes),总计13,495帧(frames),1个任务(task)。数据文件大小约100 MB,视频文件大小约200 MB。
- 数据分割:训练集(train)包含100个全部片段(索引0到100)。
- 数据特征:
action:动作数据,4维浮点数(float32)。observation.state:观测状态,13维浮点数(float32)。observation.images.camera1和observation.images.camera2:两个摄像头的视频观测,分辨率为256×256像素,3通道,帧率20 FPS,编码格式为AV1。timestamp:时间戳,1维浮点数(float32)。frame_index:帧索引,1维整型(int64)。episode_index:任务片段索引,1维整型(int64)。index:全局索引,1维整型(int64)。task_index:任务索引,1维整型(int64)。
- 机器人类型:UR5e
- 仿真环境:基于Robosuite仿真环境,模拟抓取-放置(pick-place)任务,动作采用绝对末端执行器坐标系。
- 创建工具:使用LeRobot框架创建。
- 数据集可视化:可通过可视化交互页面预览,地址为 https://huggingface.co/spaces/lerobot/visualize_dataset?path=castanetnicolas/UR5e_sim_pick_place_can_100_absolute_EEF_actions
- 引用信息:暂无提供BibTeX引用格式。




