Last fall, I stopped building and focused entirely on freelancing. It was simple: save up enough to build for >9 months without needing to stress about money. Now I’ve saved €45,000 - enough for 12 months. And the question returns: what do I actually want to build? 🧵

World models are the only scalable path for robots to learn diverse tasks with fast feedback. This is the first model that actually works well enough to enable that.

ML Models for robotic manipulation. Bimanual arms for the next 10 years and humanoids in 10+ years. Europe is world leader in process engineering. What’s missing is automation for the jobs no one wants which will simultaneously make end-product costs competitive with Asia.

The moment a robot moves like a person, it stops being a machine and starts being a character.

Why doesn’t this exist:. I upload n academic papers and get:.1) Ranking of works cited in common.2) Shared authors across papers.3) Timeline of all cited work + citation counts.4) [Optional] Newer papers that cite papers in my collection – ranked by citations/month, bonus if they.

@LeCARLab @ebiyik_ @Ishika_S_ @Marc__Toussaint CMU: USC: TU Berlin:

Interested in robot learning but not sure where to start? . I found 3 university courses with online materials (links below):. 1. CMU: Introduction to Robot Learning @LeCARLab .2. USC: Robot Learning by @ebiyik_ and @Ishika_S_ .3. TU Berlin: Robot Learning by @Marc__Toussaint.

Nice to see that the @LeCARLab at CMU agrees with me that moving chess pieces with robots is hard. 🤝🏼

Just opened R Studio by accident via @raycast and felt a wave of relief - it’s been over three years since I’ve had to touch R.

This is such a cool form factor.

Inaccurately oversimplified network diagrams in papers can be so confusing.

410$?! . The SF Marathon is completely overpriced!. So I downloaded the GPS coordinates and ran it by myself today.

It’s the weeks where you intentionally decide not to work where you make the big decisions.

RT @azerkoculu: been working on this for months and it's finally out on @ProductHunt .

RT @avrecum: First it’s complete teleoperation (for which there is no real technological blocker), then you collect the actions/videos and….

RT @ThytuVDM: I may have under-estimated how much I have to learn. Time to go back to basics

I think the most expensive mistake I've made in my robotics journey so far was trying to save 50€ by buying a cheaper camera.

RT @shipfr8: Name one other program where this is what a normal day looks like. Cohort 1.f (26 Sep to 19 Dec) applications are now open, a….

Today and tomorrow are data collection power days: 465 today with a goal of 1100 new samples total. The @huggingface recording script previously took 30-40s to process each 15s recording, but I changed it to batch process and now collection is way faster (and more fun)!

We are so naive when it comes to our assumptions about how tech will be used: always assuming it will replace something we know. AI that creates video? → “It will replace movies.”. No, video generation enables completely NEW kinds of media like this one.

Some tech debt comes from doing a sloppy job on execution, some comes from a sloppy job on understanding the problem. Great execution on the wrong problem is worse in the short & long term. Rough, throw-away work to understand the problem is worth it even though it feels bad.