Two paradigms, compared honestly
Rules are precise but brittle. Learning is flexible but data-hungry. Vehicles of the course overlap both: at the end of every part you know what each approach buys you — and where it fails.
Robotics Corner presents · free interactive course
Modern Robotics & Physical AI
How robots moved from hand-written rules to networks that learn from data — and what stands between a lab demo and a reliable machine. Every lesson runs labs in your browser where tiny neural networks train in seconds, on your own screen.
9 partsfrom classical stack to humanoids
54 lessonsacross 9 parts, in the order they build
All 9 parts liveevery lesson available right now
Labs, not slidesnetworks train in your browser
Format
Learn robotics the way it is actually done
Read a concept, then immediately run it. Each lesson ends in a lab: small neural networks train live in your browser, classical algorithms execute step by step, and you can break things on purpose to see what the failure looks like.
Labs with real training
Behavior cloning, filtering, planning and kinematics run as live simulations. Networks train in milliseconds to seconds. Nothing to install: it all happens where JavaScript runs.
A story that connects
The parts build on each other: Part 0 explains the classical stack, Part 1 learns a policy from demonstrations, Part 3 scales to VLAs. The same robot task keeps recurring, one layer deeper each time.
Syllabus
Nine parts, from control to VLAs
All 54 interactive lessons are available now, in this order: each part answers the limits of the one before it.
Your progress0 / 4
No lessons match that search. Try “VLA”, “diffusion” or “humanoid”.
Who it is for
Built for engineers crossing the line
The course is written for engineers with experience in one of two fields. You need Python and the basics of machine learning. Transformers, diffusion models and the basics of control theory are covered inside the course.
From software
ML engineers moving into robotics
- kinematics, control and planning, as much as you need to work with learned models
- what makes robotics data special and how it is collected
- the problems where classical methods still beat neural networks
From robotics
Roboticists picking up machine learning
- a step-by-step path from behavior cloning to VLAs and world models
- how π0, GR00T and Gemini Robotics are built
- ways to combine learned models with the classical stack
FAQ
Straight answers
What do I need before starting?
Python and basic machine learning vocabulary: gradient descent, a loss function, a small neural network. The classical robotics part assumes no robotics background — we build it from scratch in Part 0.
Which parts are available now?
All of them — 9 parts and 54 lessons: introduction, imitation learning, spatial representations and language, foundation models, data, RL, humanoids, world models, and product. Every lesson has interactive labs, a quiz and a practice challenge.
How do the labs actually work?
Every lab is a real computation, not a video: tiny neural networks (hundreds to tens of thousands of weights) train with gradient descent right in your browser, and classical algorithms execute step by step on canvas. Nothing leaves your machine.
Do I need a GPU or any paid tool?
No. This is a free, browser-only course. The networks are small on purpose: they are small enough to train interactively, and big enough to exhibit real failure modes.
What hardware does the course discuss?
Lessons are hardware-agnostic: the concepts run on a $100 arm as well as a humanoid. We discuss real platforms where it matters — actuation, teleoperation rigs, humanoid locomotion.