robocasa_20260430T030150Z_full_run_spicy_marinade
收藏资源简介:
RoboCasa Trajectories Single 数据集是一个用于机器人轨迹记录的数据集,每个事件对应一行数据,包含事件级别的JSON数据(如adapted_trajectory、original_trajectory和execution_metadata),以及步骤级别的序列数据(如step_index、tool_name、tool_args、robot_idx和success)。此外,数据集还包括多个图像列,如room_view、top_view、map、agentview_center、agentview_left、agentview_right和wrist。数据集总共有3000行,对应3000个事件,平均每个事件有53.4个步骤,适用于SpicyMarinade任务。数据可通过HuggingFace的load_dataset函数加载。相机渲染以JPEG格式存储,地图以PNG格式存储,但MP4视频和调试用的初始相机帧不包含在内。
The RoboCasa Trajectories Single dataset contains one row of data per RoboCasa trajectory/event. Each row represents a trajectory/event, with event-level JSON data stored inline, including adapted_trajectory, original_trajectory, and execution_metadata. Step-level data is stored in aligned sequence columns, including step_index, tool_name, tool_args, robot_idx, and success. Additionally, the dataset includes multiple image columns such as room_view, top_view, map, agentview_center, agentview_left, agentview_right, and wrist. The dataset has a total of 3000 rows, corresponding to 3000 events, with an average of 53.4 steps per event. It is suitable for SpicyMarinade tasks. Data loading can be achieved via HuggingFaces load_dataset function. Note that camera renderings are stored in JPEG format, maps in PNG format, and MP4 videos and initial camera frames for debugging are not included in the dataset.
根据您提供的数据集详情页面地址和README文件内容,我为您总结了以下关键信息:
RoboCasa Trajectories Single 数据集概述
基本信息
- 数据集名称:RoboCasa Trajectories Single
- 来源地址:https://huggingface.co/datasets/DorianAtSchool/robocasa_20260430T030150Z_full_run_spicy_marinade
- 数据集行数:3000
- 轨迹/情节数量:3000
- 任务类型:SpicyMarinade
- 每情节平均步数:53.4
数据结构
数据集每一行代表一个轨迹/情节,包含以下内容:
情节级数据(内联JSON存储)
adapted_trajectory:调整后的轨迹original_trajectory:原始轨迹execution_metadata:执行元数据
步骤级数据(对齐序列列)
step_index:步骤索引tool_name:工具名称tool_args:工具参数robot_idx:机器人索引success:是否成功
图像列(内联展示)
room_view:房间视图top_view:俯视图map:地图(PNG格式)agentview_center:智能体中心视角agentview_left:智能体左侧视角agentview_right:智能体右侧视角wrist:手腕视角
加载方式
使用以下Python代码加载数据集:
python from datasets import load_dataset
ds = load_dataset("DorianAtSchool/robocasa_20260430T030150Z_full_run_spicy_marinade", split="train")
注意事项
- 相机渲染图像存储为JPEG格式,地图保留为PNG格式
- 数据集中不包含MP4视频
- 调试阶段的
initial和pre_initial_state相机帧不包含在数据集中 - 行粒度:轨迹级别
- 该布局在
load_dataset()下可完整加载,但嵌套的序列列在Hugging Face表格查看器中不太友好




