Establishes the object of the business: what physical AI data is, which sensor modalities produce it, and how it is captured in the real world, including teleoperation. This is the prerequisite layer for every later commercial and legal lesson.
免費開始,取得私人副本並獨立記錄你的學習進度。
你的私人副本會從這個已發布版本開始。可選練習和閱讀進度會保持私密,並與發布者分開。
開始免費課程不會消耗市場點數(MC)或 AI 額度單位(AU)。
公開課程是探索學習內容的入口。閱讀預覽不代表掌握度、準備度或錄用結果。
預覽
可在這裡閱讀完整課程。閱讀不會建立私人副本、儲存進度或記錄學習證據。
預覽材料
Physical AI data is a time-aligned record of what an embodied system sensed and did while interacting with the physical world. A record counts as physical AI data when it satisfies three criteria at once.
The three criteria are a package. A weather station has sensors and a physical location, but it is not an embodied system acting in the world, so its readings are environmental data, not physical AI data. A scanned invoice has a digital source but no body, no sensor stream, and no action context.
比較
| 維度 | Warehouse robot pick episode | Scanned PDF invoice |
|---|
| Embodied source | A physical robot arm in a real warehouse | None — a document produced by office software |
|---|---|---|
| Sensor stream | Camera frames, joint encoder readings, gripper force | None — text and layout only |
| Action or state context | Commanded pick action, arm pose, task phase | None — no action or state is recorded |
| Verdict | Physical AI data | Not physical AI data |