Physical AI HandbookSeven Learning Paths for Embodied Intelligence

From probabilistic objects and action distributions to closed-loop operation on real robots—an open curriculum designed to be readable, reproducible, and continuously updated.

Start Reading

Demo Data

Start from the Objects

Distinguish among policies, world models, value functions, planning, control, and data engines to avoid being guided by model names.

Equations Must Be Readable

Every key equation is accompanied by an explanation of how to read it, its derivation, and its meaning in the robot closed loop, along with a minimal reproducible experiment whenever possible.

Integrate Evidence into the System

Course content, paper references, evaluation protocols, and demo data schemas are managed separately with full version traceability.

This Is Not a Linear Course

Direct policies, world models, value learning, hierarchical planning, data representations, control, and systems engineering address different problems within the same physical closed loop. Begin with the common foundations, then choose a primary path; cross-links throughout the chapters will indicate which layer to study next.

Who This Is For

This handbook is intended for readers with university-level knowledge of probability and statistics, linear algebra, and foundational deep learning who want to truly understand the boundaries among robot foundation models, world models, reinforcement learning, control, and data engineering.

Sources and Updates

The main content is maintained in the ArcheBase Feishu course catalog, and this repository generates the public reading version through synchronization scripts. Each release records the source document revision number and synchronization time. Unauthorized internal materials, customer data, and personal information are never included on the public site.