Robotics is almost solved! Pun intended. Sunday published a "Solve" standard. They declared the scope (garment types, lighting, surfaces, starting configurations) and adaptation cost (zero—no fine-tuning per home) before evaluating. The same model weights run everywhere. This might be the differentiation between a good demo and reliable work of... robots. They presented ACT-2, a robot foundation model that folds laundry at 99.1% success across unseen homes. Different houses, lighting, bed types, garment sizes. According to them scaling pre-training closes the generalization gap. Train on massive, high-quality human data. When you do that right, what you learn in-house actually transfers to unseen homes. Even wilder is that the model learns new folding techniques from a single example and generalizes them to new environments. Memo goes to homes later this year. Would you adopt this cutie? 🥹 Read more here: Great work Tony Zhao and team! 🧡 ~~ ♻️ Join the weekly robotics newsletter, and never miss any news →show more

@sundayrobotics put it in a hoarders house and put it on camera so we can see the robot have a panic attack

@sundayrobotics This is groundbreaking, only caveat I’d make is that the “new environments” aren’t messy enough. When they start deploying in homes this will change... But anyway the generalization they showed was remarkable

@sundayrobotics 99.1% laundry folding success across unseen homes is genuinely impressive

@sundayrobotics А вы хорошо подумали о том, какое влияние, этот "малыш", окажет на человеческие нейросети?

@sundayrobotics the "declare your scope before evaluating" move is doing a lot of work here. most robot demos skip that part entirely and hope nobody notices

@sundayrobotics From the very bottom of the skill curve: it's solved enough that a non-technical mom trained a dual-arm kit at home this week. Not solved enough that it survives me nudging the overhead camera 5 cm. That gap is exactly where the next standard lives 😄

@sundayrobotics The key to generalisation all along was to call solved solved.

@sundayrobotics Yes, I would adopt Memo. Personally from physiological point of view, I’d be far more comfortable having this kind of human-machine hybrid at home, much less intimidating than robots look exactly like humans that might kidnap my kids while I’m sleeping. 😜😁

@sundayrobotics Sick

@sundayrobotics I want!

@sundayrobotics

@sundayrobotics 这牛啊

@sundayrobotics trade school earn chart wages plus the patience is the data the trades earn every quarter

@sundayrobotics 🧐 @crux_capital_ @damnang2

@sundayrobotics "Train on massive, high-quality human data." You know specifically they didn't use synthetic data??

@sundayrobotics hold your horses, still far away yet, but in the right path

@sundayrobotics We see similar transfer when our sim randomizes garment physics and lighting across thousands of episodes. That single-example adaptation is what domain randomization really unlocks at scale.

The reason they make the robot human shape, is that they are trying to make the robot multi tasks oriented. That means it, they want to make the robot to do quite a lot different tasks in one type of robot. but is that the right choice? For example, in this case, is it really necessary for a robot to vacuum the floor with a vacuumer which is designed for human?