How to read a lesson
Every lesson has the same shape, so you always know where the theory, the code and the evidence are.
Metadata
Section titled “Metadata”| Row | Meaning |
|---|---|
| Part | The part the lesson belongs to |
| Run type | CPU runs on a Mac or Colab CPU · GPU needs a Colab GPU · ART uses results we ship (logs, checkpoints) and runs on CPU · HW needs the real SO-101 |
| Prerequisites | Earlier lessons of this book you need |
| Learn first | Lessons of AI Engineering from Scratch that cover the general background (math, ML, deep learning, RL) |
| Time | Reading and running, roughly |
| Code | The experiment script (and notebook, if any) |
| Status | ✅ done · 📝 draft · 🚧 coming soon |
Sections
Section titled “Sections”| Section | What it gives you |
|---|---|
| The problem | One concrete failure or question the lesson resolves |
| Core idea | The minimum theory: definitions and a few equations |
| Experiment | Predict, run, observe. Short code, then a table of measured numbers |
| Break it | One parameter pushed until the method fails, and why |
| On the real robot | What changes on hardware, with sourced facts only |
| Check yourself | Three exercises, easy to hard, with answers |
| Takeaways and sources | Three bullets, the next lesson, primary sources |
Every experiment is a sweep
Section titled “Every experiment is a sweep”One parameter changes and one quantity is measured, so the result is a table, not a single anecdote. The exact output of each run is saved next to its script as output.txt; the numbers in the lesson come from there.