Robot Learning 2026 Main Lectures — ETH Zürich (Oier Mees)

TL;DR

YouTube playlist of the main lectures from ETH Zürich’s Spring 2026 course “Robot Learning: From Fundamentals to Foundation Models” (course code 263-5911-00L, 5 ECTS), taught by Oier Mees as an external lecturer (his day job: lead research scientist for robot learning at Microsoft, previously postdoc with Sergey Levine at UC Berkeley; co-creator of Octo). The playlist’s canonical entry is Lecture 1 (Introduction to Robot Learning, posted ~one month before filing); the full series walks from classical imitation learning, MDPs, and policy optimization through to Vision-Language-Action (VLA) models and foundation models for robotics. Posted to the thread as a correction — the original thread link was the guest lectures playlist; this is the main lectures playlist that contains the core syllabus content.

Key claims

The course is structured as weekly sessions each combining a main lecture on a core topic with a student-led paper discussion, plus occasional short guest lectures from external experts [course page intro].
The syllabus progresses from fundamentals (imitation learning, reinforcement learning, policy optimization) to advanced topics including VLA models and foundation models for robotics [VVZ catalog description].
Lecture 1 (the playlist’s canonical entry) is the Introduction to Robot Learning [Lecture 1 title]; Lecture 3 covers Imitation Learning [Lecture 3 title].
Practicals are organized as four homework sets — PyTorch tutorial, robot control/MDPs, imitation learning, reinforcement learning — released through the mees-robot-learning-course/ethz-course-2026 GitHub repo [course GitHub].

Method

The artifact is a curated YouTube playlist (PLPU18BnWYUZJx3_d901-GD6BGpeWwE2vx, titled “Main Lectures - ETHZ Robot Learning Course 2026”) containing the main lectures delivered by Oier Mees. The course’s parallel guest lectures playlist (separate, and what the upstream thread message linked) hosts external speakers — Ted Xiao on “Three Eras of Robot Learning,” Cheng Chi on “Robotics Beyond Algorithms,” Lucas Beyer on “Vision in the Age of LLMs,” and others — with each lecture posted as an individual video. The course page lives at cvg.ethz.ch/lectures/Robot-Learning/; the practicals repo lives at github.com/mees-robot-learning-course/ethz-course-2026. The filing canonical_source is set to Lecture 1’s standalone video URL (so the page can deduplicate against future re-shares of any single-video form), but the artifact-of-interest is the full playlist.

Results

Not applicable — this is a teaching artifact, not a research result. Notable surrounding signal: the GitHub practicals repo had 398 stars / 124 forks at filing time, indicating substantial out-of-Zurich uptake.

Why it’s interesting

This is the closest current thing to a canonical “what does a modern robot-learning curriculum look like” reference for the team, and the syllabus topics map directly onto active wiki coverage: imitation learning + RL + policy optimization feed into Reasoning RL and the team’s VLA tracking under VLA Models (where the field is actively debating action-pretraining vs. unified-VLM vs. clean-teleop recipes — see π*0.6: a VLA That Learns From Experience (RECAP), Embodied-R1.5: Evolving Physical Intelligence via Embodied Foundation Models, Spirit-v1.5: Clean Data Is the Enemy of Great Robot Foundation Models). The foundation-models-for-robotics segments connect to World Foundation Models — particularly recent material the team has been chasing under that concept (e.g. Yann LeCun: World Models — Enabling the next AI revolution (ETH Frontiers of Embodied AI), a guest lecture in a separate ETH series). Pairing this main-lectures playlist with the guest-lectures playlist (linked in the parent thread) gives the same coverage the wiki is otherwise reconstructing one paper at a time.