jacob-valdez/synthux-economy-r3-w021
收藏资源简介:
SynthUX计算机使用数据集(经济模拟r3)是一个基于模拟公司环境的视觉计算机使用轨迹数据集。它通过SynthUX生成,记录员工在模拟设备会话中的操作过程,包括将自然语言目标扩展为节点树,终端节点通过低级鼠标/键盘输入驱动真实桌面模拟应用程序(如终端、笔记、VS Code、浏览器、Slack/Teams、邮件、表格、幻灯片等),并捕获观察轨迹作为屏幕视频及每个节点的帧。数据集支持应用原生多应用程序执行(无工作台覆盖),提供每参与者内容多样性、跨操作系统消息传递和大规模多公司经济模拟。每个经济模拟作为独立的数据集存储库发布。内容包含数据记录文件(data.jsonl)、每个轨迹的截图(按节点ID组织)和完整屏幕录制视频。记录模式涵盖轨迹ID、参与者信息(如演员ID、公司、角色ID、职位)、模拟器环境(macos-web-next、windows-web-next、browser-os)、目标、节点树、密集低级输入事件、因果模拟器状态观察、对齐链接以及媒体引用。本构建版本包含36个轨迹、978个帧、涉及6家公司,环境分布为browser-os(17个)、macos-web-next(10个)和windows-web-next(9个)。数据生成使用SynthUX工具,观察基于因果模拟器状态(突变通道为低级输入),捕获过程中未使用高级应用程序突变。
The SynthUX Computer-Use Dataset — economy sim r3 is a dataset of grounded visual computer-use trajectories generated by SynthUX in a simulated company environment. Each record represents one workers device session, where a natural-language goal expands into a node tree, terminal nodes drive real desktop-simulator applications (such as Terminal, Notes, VS Code, Browser, Slack/Teams, Mail, Sheets, Slides, etc.) through low-level mouse/keyboard input, and the observed trajectory is recorded as a screen video plus per-node frames. The dataset features app-native multi-application execution (without a workbench overlay), per-actor content diversity, cross-OS messaging, and large multi-company economies. Each economy simulation is published as its own dataset repository. Contents include a data record file (data.jsonl), screenshots per trajectory organized by node IDs, and full screen recording videos. The record schema covers trajectory ID, participant information (actor ID, company, role ID, title), simulator environment (macos-web-next, windows-web-next, browser-os), goal, node tree, dense low-level input events, causal simulator state observations, alignment links, and media references. This build includes 36 trajectories, 978 frames, and 6 companies, with environment distribution: browser-os (17), macos-web-next (10), and windows-web-next (9). The data is generated using SynthUX, with observations based on causal simulator states (mutation channel = low-level input) and no high-level app mutation used during capture.




