The book can be read front to back, which is the strongest path, or by route. Engineers can begin with the setup chapter, simulation stack, and deployment chapters. Researchers can jump to VLA models, world models, humanoids, and evaluation after reading the foundations. Instructors can use the course design chapter and appendices to build an advanced undergraduate or graduate offering.
For a fast builder path, read Chapters 0, 1, 3, 9, 10, 11, 21, 22, 23, 34, 52, and 55. For a research path, read Chapters 4 to 8, 14 to 25, 34 to 41, 46, 52, 54, and 58.
The fastest-moving systems in this book, including V-JEPA 2, TD-MPC2, ALOHA 2 and Mobile ALOHA, RoboMIND, AgiBot World, recent vision-language-action models, flow-matching action experts, and cross-embodiment foundation models, are covered from preprints and company or lab reports rather than from settled literature. Treat their specifics as a snapshot, and validate before relying on any number: terminology is still shifting, reported benchmarks are often not directly comparable across action spaces and simulators, data and model availability change, and claims about scaling and generalization are frequently preliminary. Where this book cites such results, read them as directional evidence, and check the primary source and its evaluation conditions before quoting a figure.