synthux-economy-r5-w063
收藏资源简介:
SynthUX Computer-Use Dataset — economy sim r5 是一个由SynthUX生成的基于视觉的计算机使用轨迹数据集,专门用于模拟公司环境中员工的操作行为。数据集的核心是记录员工在模拟公司内的完整工作会话:每个会话从一个自然语言目标开始,该目标被扩展为节点树,终端节点通过低级鼠标和键盘输入驱动真实的桌面模拟器应用程序(包括终端、笔记、VS Code、浏览器、Slack/Teams、邮件、表格、幻灯片等)。观测到的轨迹被记录为屏幕视频以及每个终端节点对应的时间点截图。数据集强调应用原生的多应用程序执行(无工作台覆盖)、每个参与者内容的多样性、跨操作系统的消息传递以及大型多公司经济模拟。每个经济模拟都作为独立的数据集仓库发布。数据集内容包含:1) `data.jsonl`文件,每条记录对应一个渲染的轨迹;2) `screenshots/<trajectory>/<node_id>.png`目录,为每个终端节点在其最后输入事件时刻截取的帧图像;3) `videos/<trajectory>.webm`目录,完整的屏幕录制视频。记录模式包含轨迹ID、参与者与组织信息(演员ID、公司、角色ID、职位)、模拟器环境、自然语言目标、节点树结构、密集的低级输入事件序列、因果模拟器状态观察结果、观察与事件的节点对齐信息以及媒体引用。本版本(r5)的具体统计为:共36条轨迹,对应36个视频和879帧截图,涉及6家公司,模拟器环境分布为browser-os(8条)、macos-web-next(10条)、windows-web-next(18条)。数据由SynthUX工具生成,观察基于因果机制(突变通道为低级输入),在捕获过程中未使用高级应用程序突变。
SynthUX Computer-Use Dataset — economy sim r5 is a vision-based computer usage trajectory dataset generated by SynthUX, specifically designed to simulate employee operational behaviors in corporate environments. The core of this dataset is to record full work sessions of employees within simulated corporate environments: Each session starts with a natural language goal, which is expanded into a node tree, where terminal nodes drive real desktop simulator applications (including terminal, Notes, VS Code, browsers, Slack/Teams, email, spreadsheets, presentations, etc.) via low-level mouse and keyboard inputs. Observed trajectories are recorded as screen videos and timestamped screenshots corresponding to each terminal node. This dataset emphasizes native multi-application execution (no workspace overlay), diversity of content per participant, cross-operating system messaging, and large-scale multi-corporate economic simulations. Each economic simulation is released as an independent dataset repository. The dataset includes the following components: 1) `data.jsonl` file, where each line corresponds to one rendered trajectory; 2) `screenshots/<trajectory>/<node_id>.png` directory storing frame images captured at the moment of the final input event for each terminal node; 3) `videos/<trajectory>.webm` directory containing full screen recording videos. The record schema includes trajectory ID, participant and organizational information (actor ID, company, role ID, job title), simulator environment, natural language goal, node tree structure, dense low-level input event sequences, causal simulator state observations, node alignment information between observations and events, and media references. Specific statistics for this version (r5) are as follows: A total of 36 trajectories, corresponding to 36 videos and 879 screenshot frames, involving 6 companies. The distribution of simulator environments is: browser-os (8 trajectories), macos-web-next (10 trajectories), windows-web-next (18 trajectories). The dataset is generated by the SynthUX tool. Observations are based on causal mechanisms (with low-level inputs as the mutation channel), and no high-level application mutations were used during the capture process.
数据集概述
- 数据集名称:SynthUX Computer-Use Dataset — economy sim r5
- 许可协议:Apache-2.0
- 任务类别:其他(other)
- 标签:computer-use, gui-agents, synthetic, multi-application
数据集内容
该数据集包含模拟公司中工作人员的桌面操作轨迹,每条记录代表一个完整的设备会话。内容通过 SynthUX 生成,包括以下文件:
data.jsonl:每条渲染轨迹的记录。screenshots/<trajectory>/<node_id>.png:每个终端节点在其最后一次输入事件时的帧截图。videos/<trajectory>.webm:完整的屏幕录制视频。
记录字段说明
| 字段 | 含义 |
|---|---|
trajectory_id |
唯一标识(<actor>__<env>) |
actor_id, company, role_id, title |
操作者所属组织及身份信息 |
env |
模拟器环境:macos-web-next / windows-web-next / browser-os |
goal |
会话的自然语言目标 |
tree.nodes |
操作者的终端节点(包含 surface, action, target, args) |
trajectory.input_events |
密集的低级鼠标/键盘/脚本事件(含时间戳 t_ms) |
trajectory.observations |
每个节点后的因果模拟器状态 |
trajectory.alignment |
将观测值和输入事件范围链接到节点 ID |
media.video |
屏幕录制(webm 格式) |
media.frames |
每个节点的 PNG 帧(含 node_id, t_ms, blob_ref) |
数据统计(当前版本)
- 轨迹数:36 条
- 视频数:36 个,帧数:879 帧
- 公司数:6 个
- 环境分布:browser-os(8 条),macos-web-next(10 条),windows-web-next(18 条)
数据来源
数据集通过 SynthUX(synthux.multicompany.day.v1)生成。观测数据是因果性的(突变通道为低级输入),在捕获过程中未使用高级应用突变。





