Learning Robot Learning

Concepts, pictures, and code for ETH Zürich’s Robot Learning course

This site is an independent, unofficial companion to ETH Zürich’s Robot Learning: From Fundamentals to Foundation Models, taught by Oier Mees in Spring 2026. The course page has the lecture recordings, slides, and homework.

Each course week has a page with the big picture, intuitions, the key equations explained term by term, worked NumPy examples, interactive figures, a short animation of the week’s key idea, a PyTorch sketch of the core algorithm, and summaries of the papers and the guest talk.

The course in one picture

The course moves from the robot itself to the models that are starting to control it (Figure 1). The first weeks set up the robot and the language for its decisions, then the two ways a robot can learn behaviour: by copying demonstrations and by trial and error. The middle of the course is the architectures of the foundation-model era: generative models, transformers and world models. The last weeks put them together into generalist, reasoning policies, and ask what is still missing.

arc f Foundations  · Weeks 1–2 the robot, control, MDPs l Learning behaviour  · Weeks 3–5 from demonstrations and from reward f->l learn instead of program a Modern architectures  · Weeks 6–8 generative models, transformers, world models l->a more expressive policies p Putting it together  · Weeks 9–10 generalist policies, reasoning a->p scale and combine o Looking ahead  · Weeks 11–12 open problems, guest talks p->o what is missing
Figure 1: How the parts of the course build on each other. Each arrow says what the next part adds.

Weeks

Week The question it answers
1 · Introduction to Robot Learning Why can’t we simply program robots for our homes, and what does learning from data change?
2 · Robot Control & MDPs How does a goal become motor commands, and how do we describe the decisions above that?
Week The question it answers
3 · Imitation Learning How do demonstrations become a policy, and why does copying break down?
4 · Reinforcement Learning I How can a robot learn from rewards alone, with and without a model? (values, Q-learning, DQN)
5 · Reinforcement Learning II How do you follow the gradient of reward through a world you can’t differentiate? (policy gradients, PPO, SAC)
Week The question it answers
6 · Generative Models How can a policy represent a complicated distribution over actions? (diffusion, flow matching)
7 · Sequence Modeling and Transformers How do observations, words and actions become one sequence a transformer learns from?
8 · World Models How do you learn what your actions will do, and use it?
Week The question it answers
9 · Generalist Robot Policies How do you train one policy for many robots, and tell whether it works?
10 · Embodied Reasoning and Test-time Scaling Can a robot that thinks harder act better, and when is thinking worth its cost?
Week The question it answers
11 · Frontiers & Open Problems What is still unsolved, and which ideas point towards solutions?
12 · Guest Lectures How do the pieces fit together? (the course reviewed in one table; the guest talk is not available online)

Reading paths

You don’t have to read in order. Some routes through the site, each a chain of weeks that build on one another:

Table 1: Suggested orders for common goals.
If you want Read
A quick tour the Big picture and Intuition of every week, then 12
Learning from demonstrations 3 → 6 → 7 → 9
Reinforcement learning 2 → 4 → 5 → 11
Foundation models for robots 7 → 9 → 10 → 11
Models of the world 2 → 4 → 8

How to use this site

Each week’s page is laid out the same way:

  1. Big picture: where the week fits, its key ideas and its vocabulary.
  2. Intuition: a concrete story and pictures before any math, with ▶ links to the moment in the lecture video where each idea is taught.
  3. Key concepts & formulae: the equations, in the course’s notation, each explained in words and term by term.
  4. Worked examples: the ideas applied to real numbers.
  5. Interactive: figures you can change.
  6. Algorithm & code: the NumPy code that produced the page’s figures, beside a PyTorch sketch of how you would write it at scale.
  7. Papers and Guest spotlight: original summaries of the papers discussed in class and of the guest talk.
  8. Solved problems, Recap and Cheat sheet: try each problem before opening its solution.

Big picture and Intuition are written to stand alone: reading just those two sections of every week gives a tour of the whole course. Week 12 is shorter, since it has no lecture.

The pages never reproduce the course’s slides or homework solutions; the course page is the place for those.

Three reference pages collect material from across the site:

  • Papers: every paper card, grouped by week.
  • Glossary: every symbol and term, grouped by week, with a table of symbols that change meaning.
  • Cheat sheets: every week’s key equations in one place.

For the decision-making theory underneath Weeks 2 to 5, the sibling ADM Companion goes deeper.