leosltl/Android-Control

Android Control是一个用于训练和评估Android设备控制代理的数据集。它包含15,283个演示，覆盖833个Android应用和40个应用类别，包含14,548个独特任务。每个演示包括：截图（PNG编码）、无障碍树（序列化的协议缓冲区格式）、自然语言指令（高级目标和低级逐步指令）以及动作（如点击、滚动、输入文本等）。数据以TFRecord文件格式存储，并采用GZIP压缩。每个记录包含演示的唯一标识符、高级任务描述、低级动作描述、屏幕截图、无障碍树和动作序列。

Android Control is a dataset for training and evaluating Android device control agents. It contains 15,283 demonstrations across 833 Android apps covering 40 app categories, with 14,548 unique tasks. Each demonstration includes: Screenshots (PNG-encoded), Accessibility trees (serialized protocol buffer format), Natural language instructions — both high-level goals and low-level step-by-step instructions, Actions — `click`, `scroll`, `input_text`, `open_app`, `navigate_home`, `navigate_back`, `long_press`, `wait`. The data is stored as TFRecord files with GZIP compression. Each record contains: `episode_id`, `goal`, `step_instructions`, `screenshots`, `accessibility_trees`, and `actions`.