The companion website for the forthcoming monograph by the Physical Intelligence Team, Shanghai AI Laboratory.
The monograph develops a unified view of Physical AI: artificial systems that perceive, model, and act upon the physical world through continuous feedback under real-world constraints. It connects foundations in physics, control, cognition, learning, and reasoning with perception, data, world models, embodied foundation models, physical agents, and hardware–software systems.
The forthcoming monograph contains 12 chapters in three parts:
| Part | Chapters | Scope |
|---|---|---|
| I. Overview and Foundations | 1–2 | Definitions, boundaries, and technical foundations |
| II. Methodology | 3–9 | Perception, data, world models, manipulation, motor control, agents, and systems |
| III. Future Outlook | 10–12 | General physical intelligence, evaluation, safety, governance, and development strategies |
The website includes the book introduction, a three-part expandable table of contents, author information, companion resources, contact information, and a provisional citation. An original concept cover and a vector overview figure are included. The manuscript is not yet publicly released. The only arXiv reading link appears under Explore Intelligence in the Physical World. It currently points to an unpublished placeholder that returns 404.
Requires Node.js 20.9+ and pnpm. The pnpm version is specified in package.json.
pnpm install --frozen-lockfile
pnpm devOpen the local URL printed by Next.js.
pnpm typecheck
pnpm test:analytics
pnpm buildThe static export is generated in out/. The GitHub Pages workflow is in .github/workflows/deploy-pages.yml; it configures the deployment base path automatically.
| Content | Location |
|---|---|
| Book information, cover configuration, resource links, and citation | src/data/project-page.ts |
| Parts, chapters, sections, and page numbers | src/data/contents.json |
| arXiv destination and release switch | src/data/publication.ts |
| Click collection, persistence, and aggregation | src/lib/analytics.ts |
| Analytics dashboard | src/app/analytics/ and src/components/analytics/ |
| Unreleased manuscript (local only, excluded from Git and deployment) | reference/private-manuscript/ |
| Cover artwork and book figures | public/images/monograph/ |
| Page sections and layout | src/components/research-page/ |
| Theme and responsive styles | src/app/globals.css |
The current concept cover is in public/images/monograph/cover-v2.png. To replace it with final artwork, place the new image in the same directory and update bookCover.src, its dimensions, and alternative text in src/data/project-page.ts. The existing hover interaction and mobile placement apply to the replacement automatically.
After release, set arxivPublication.url to the official public arXiv abstract URL and set arxivPublication.published to true in src/data/publication.ts, then rebuild. Both are required to start counting. The button links directly to arXiv without an intermediate page. Keep unpublished PDFs outside public/; reference/ is ignored by Git and is not part of the site export. The manuscript extraction script reads the local copy under reference/private-manuscript/.
Unavailable resource links use href: null. Add their URLs and labels when the materials are ready. Publisher, publication date, and purchase link are configured in projectInfo. The table of contents displays one column per part on desktop and one column on smaller screens. Chapter rows expand or collapse their sections; the table of contents has no reading links. The former /read/ and /chapters/ pages have been removed.
Set contactInfo.email in src/data/project-page.ts to enable the InternRobotics email link. While it is null, the site displays an email placeholder without a fabricated address.
The bottom arXiv button displays a text counter beneath it. Before release, this counter and the dashboard stay at 0; placeholder clicks do not write local records or send analytics events. After the official link and publication switch are enabled, left clicks, keyboard activation, and middle clicks record arxiv_click events. These are button clicks, not unique readers or completed downloads.
After release, daily counts are saved in this browser only by default, under the physicalai:arxiv-clicks:published:v1 localStorage key. Earlier placeholder test records are excluded. The homepage counter and dashboard share the same data source. The dashboard shows total clicks, today's clicks, the last seven days, 7/30/90-day charts, a daily table, and CSV export. Dates use Asia/Shanghai (UTC+8). Different visitors, devices, browsers, and browser profiles are not combined. Clearing browser data removes local counts. Storage or analytics failures do not block the arXiv link.
No external analytics service is enabled by default. Two optional HTTPS endpoints are ready for future global collection and visualization:
| Configuration | Purpose |
|---|---|
NEXT_PUBLIC_ANALYTICS_EVENT_ENDPOINT |
Receives click events via POST |
NEXT_PUBLIC_ANALYTICS_SUMMARY_ENDPOINT |
Supplies the dashboard's aggregated daily counts via GET |
For local development, copy .env.example to .env.local. For GitHub Pages, set these names as repository Actions variables and rebuild. Leave both empty for browser-local analytics. A collection endpoint alone does not turn local dashboard data into global counts: the summary endpoint must also be configured. When the summary endpoint is configured, the dashboard labels its data as global and reports service failures explicitly.
The collection payload is JSON sent with a text/plain content type, so the endpoint should parse the text body as JSON:
{"event":"arxiv_click","event_id":"unique-event-id","occurred_at":"2026-10-14T08:00:00Z","source":"bottom_cta","destination":"https://arxiv.org/abs/2601.12345"}The backend should deduplicate event_id, aggregate dates in Asia/Shanghai, and return this summary shape with CORS access for the website:
{"version":1,"daily":{"2026-10-14":12},"updatedAt":"2026-10-14T08:00:00Z"}The daily count above is a schema example, not live data. No reader identity, IP address, referrer, or browser fingerprint is added to the application payload. An optional integration can listen for the physicalai:analytics browser event; its detail also indicates whether the local count was saved.
@misc{physicalai2026monograph,
title = {Physical AI: Concepts, Techniques, and Challenges},
author = {{Physical Intelligence Team, Shanghai AI Laboratory}},
year = {2026},
note = {Forthcoming; manuscript not yet publicly released}
}The citation will be updated when final publication details are available.