Guan Sun

ML Research @sundayrobotics | Prev: Google, Waymo, Amazon Robotics

[Guan Sun @Guan__SUN]

The long tail of real-world robotics looks like this: a 4-year-old deciding, on his own, to hand your robot a t-shirt.

Memo just took it naturally and started folding — my son's smile afterward says more than any eval.

Today we published how: https://t.co/tGs8y0U1Rq

[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.

[Guan Sun @Guan__SUN]

Every emergent behavior of ACT-2 started as someone's careful demonstration.

A thousand quiet co-authors. This thread is their story.

Thank you @perryzjia — and everyone behind the data engine — for the effort that made all of it possible.

[Perry Jia @perryzjia]

1/N Two years ago, we recruited our first Memory Developer off Craigslist and onboarded them in a public library.

Today, more than 1,000 Memory Developers have helped us build the data engine behind ACT-2.

Here's how we got from that library to here 🧵

[Guan Sun @Guan__SUN]

@perryzjia @sundayrobotics Thank you Perry for making happiness a pillar. Of the four, it's the one that makes the other three possible.

[Guan Sun @Guan__SUN]

Your laundry should not depend on your WiFi.

[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 🧵

[Shuran Song @SongShuran]

One year ago, I wouldn't have imagined that laundry folding would be the first "solved" task for home robots. The world is moving soooooo fast!

[Charles Qi @charles_rqi]

The four essentials of daily life are clothing, food, housing, and transportation.

Clothing may be the first category that home robots truly crack.

Congrats to Sunday Robotics for the ACT-2 launch that got 99% accuracy in the laundry folding task!

[Guan Sun @Guan__SUN]

@zhaodloke Thanks for making eval work so well for us!

[Eric Jang @ericjang11]

The money shot

[Alper Canberk @alpercanbe]

3 hours of Memo folds to relax/study to
https://t.co/BkQMW6mFjt

[Guan Sun @Guan__SUN]

@ASong408 Both hands work! What you're actually seeing is my son switching sides mid-handover — Memo adjusted on the fly.

[Guan Sun @Guan__SUN]

The long tail of real-world robotics looks like this: a 4-year-old deciding, on his own, to hand your robot a t-shirt.

Memo just took it naturally and started folding — my son's smile afterward says more than any eval.

Today we published how: https://t.co/tGs8y0U1Rq

[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.

[Guan Sun @Guan__SUN]

@ShiweiShengR Impressive earns a retweet. Useful earns a place in your home.

[Guan Sun @Guan__SUN]

@alpercanbe Memo learns from every mistake exactly once — the most emotionally mature coworker I've ever had.

[Guan Sun @Guan__SUN]

@JunyaoShi Five-minute fine-tuning means the model now iterates faster than my PRs get reviewed.
For the first time in robotics history, the robot is not the bottleneck. I am 🤣

[Guan Sun @Guan__SUN]

@nishantsdesai Last year we tried to be smarter than the data. This year we let the data be smarter than us.

Still wrapping my head around how much that one shift unlocked.

[Guan Sun @Guan__SUN]

You know your model is generalizing well when a 4-year-old immediately treats it like a friend. Memo and my son just had a natural, unscripted interaction that genuinely surprised me. More to share soon... stay tuned.

[Tony Zhao @tonyzzhao]

Just saw the craziest emergent behavior I’ve ever seen in robotics. The team is absolutely losing it. We haven’t posted in a while… should we write it up?