Yajvan Ravan

ML @sundayrobotics, MIT ‘25

[Tony Zhao @tonyzzhao]

the comment section on our 3 hour (15 hours wall clock) memo folding video is hilarious

[Joe Harris @_joe_harris_]

the difference between a robotics lab and a robotics team is the metrics they track:

research-strong teams:
– benchmark scores
– demo success rate
– model architecture choices

ops-strong teams:
– time from failure to root cause
– cost per deployment
– how many runs need human review
– was deployment #4 faster than #3

the second list builds the future into existence

[Yajvan Ravan @Yajvan_Ravan]

yet another example of why robotics is a systems problem. Every single component from data -> manufacturing -> hardware -> electronics -> software -> models must move in sync.

Nature spent hundreds of millions of years evolving locomotion, and millions more refining dexterity and tool use. Human language emerged on a far shorter timescale.

[Cheng Chi @chichengcc]

We just hit a weird milestone: our model became more reliable than your average home WiFi.

Just like everybody else, we thought cloud inference was the obvious choice. Yet 2 days into the ACT-2 eval, our mind completely changed.

If our hero @ArpitKalla didn’t cook, this video wouldn’t exist 🧵

[Yajvan Ravan @Yajvan_Ravan]

@sundayrobotics @JakeRoggenbuck 😭😭😭devious slander

[Yajvan Ravan @Yajvan_Ravan]

@minazakharySF Little did you know that’s what kept us focused

[Yajvan Ravan @Yajvan_Ravan]

@alpercanbe Genuinely so calming

[Yajvan Ravan @Yajvan_Ravan]

> be 2 years old
> can’t turn my clothes inside out
> cry
> wait 21 years
> see Memo turn pants inside out
> cry (tears of joy)

[Tony Zhao @tonyzzhao]

Introducing ACT-2 Preview

The first robotics model to unify broad generalization with high reliability. A single fine-tuning example can teach Memo a new behavior that generalizes.

Zero shot, real unseen homes, 99% success rate.

[Yajvan Ravan @Yajvan_Ravan]

@Guan__SUN Can’t wait for my future kids to grow up around Memo