Learning Robot Learning

An independent, unofficial companion site for ETH Zürich's Robot Learning: From Fundamentals to Foundation Models (Oier Mees, Spring 2026). Course page: https://cvg.ethz.ch/lectures/Robot-Learning/.

Site: https://laventura.github.io/learn-robot-learning/ (not published yet)

Setup

Install Quarto 1.9+ and uv. Then:

uv sync                          # Python 3.12 environment; installs rlc/ editable, includes torch (CPU)
uv run quarto check jupyter      # confirm Quarto uses .venv's Python

Commands

uv run quarto preview                               # live preview
uv run quarto render                                # build the site into _site/
uv run pytest                                       # tests
uv run python tools/check_snippets.py [--week 3]    # run the PyTorch snippets
uv run python tools/check_links.py                  # data files + "▶ mm:ss" watch links
uv run python tools/check_week.py 6                 # count a week page against the checklist
uv run python tools/check_freeze.py                 # cached results match the sources
uv run quarto publish gh-pages                      # deploy

Gathering sources (into the git-ignored refs/):

uvx yt-dlp --write-auto-subs --sub-langs en --skip-download --sub-format vtt \
    -o "refs/transcripts/w03-lecture.%(ext)s" "https://youtu.be/<id>"
uvx yt-dlp --print duration --skip-download "https://youtu.be/<id>"     # seconds -> weeks.yml
uv run python tools/clean_transcript.py refs/transcripts/w03-lecture.en.vtt
pdftotext -layout refs/<file>.pdf refs/<file>.txt

Always run Quarto through uv run, so it uses the project's Python.

Dev-machine tools

Not needed to build the site, only to gather sources and check pages.

Tool	Use
poppler (brew install poppler)	pdftotext, pdftoppm: read slide and paper PDFs
yt-dlp, run as uvx yt-dlp	Download YouTube auto-captions and video metadata
Chromium	Headless screenshots of rendered pages
ffmpeg (brew install ffmpeg)	Inspect rendered animations

Layout

• weeks/: one page per course week (wNN-slug.qmd); weeks/data/ holds JSON for interactive figures, weeks/media/ holds animation clips.
• data/weeks.yml, data/papers.yml: one source of truth for each week's videos, slides, guest, homework and papers.
• rlc/: Python package (tiny NumPy networks in nn.py, figure style, page helpers, JSON export).
• templates/_week.qmd: skeleton for a week page.
• tools/check_snippets.py: runs the PyTorch code shown on the pages.
• tools/check_links.py: validates the data files and every timestamped watch link.
• tools/check_week.py: counts a week page's sections, figures, equations, links and problems against the definition of done.
• tools/clean_transcript.py: turns YouTube auto-captions into readable text with [mm:ss] markers.
• tools/check_freeze.py: finds pages whose cached results no longer match their source.
• _freeze/: cached execution results, committed so the site rebuilds without running code.
• index.qmd, papers.qmd, glossary.qmd, cheatsheets.qmd (every week's cheat sheet, rebuilt on each render).

License

• Prose, figures, diagrams and problems: CC BY-NC 4.0; see LICENSE-content.md.
• Code (rlc/, tools/, tests/, and the code shown on the pages): MIT; see LICENSE.

The course materials belong to their authors; this site does not reproduce slides, transcripts or paper figures.