"Any pattern or structure that can be found or generated in nature can be efficiently discovered and modelled by a classical learning algorithm."
— Demis Hassabis, Nobel Lecture, 2024
For 40+ years, building a robot that could rally with an elite human table tennis player at full speed was an unsolved problem. Sony AI's Ace research project set out to change that—and the results are now accepted for publication in @Nature and featured on the cover.
It only needs a few crazy ones to fix a continent… Let's be crazy… I got something to announce:
We're launching PROTOTYPE: a new fund, fully focused on Europe.
A small fund that punches way above it's size.
With it we back what Europe is world-class at: robotics, automation, manufacturing, and anything that requires hard engineering.
First check. First round. As early as it gets.
Europe invented industry. We're the birthplace of precision manufacturing. The second largest manufacturing hub on the planet. Leaders in automation and robotics.
And yet…
We sell out our best tech to China. We export our best founders & most of our investment money to the US. That's insanity.
What should we do instead?
Build the next trillion euro companies in robotics, manufacturing, automation right here: in Europe. Showcase to young founders what is possible and change the system around them where needed.
What we will do:
→ Publish all our fund updates and build in public: https://t.co/JROaYikB2Y
→ Showcase Europe’s Most Ambitious Startups on Youtube: https://t.co/l3xVkjtbox
→ Launch & support more projects like @euinc_petition. Enough talking about Europe. Time to build it: Startups, makerspaces, student clubs, and much more.
→ And most importantly, invest into the best founders in Europe. Same model as our previous funds that are in their top 1-5% cohorts worldwide.
We won’t do a VC fund in the classic sense.
This will be a community of hyper-ambitious people who want to actively change Europe for the better.
It only needs a few crazy ones to fix a continent… Let's be crazy.
Check out https://t.co/h2lPmypoaM!
Simulation has been a central theme of my PhD, but one thing has always troubled me (and many colleagues): complex physics and rich sensing are incredibly hard to simulate.
In this work, we explore a different path: learn a simplified task in simulation, then use that sim-learned skill to collect real data for a much harder task - one that’s even challenging for teleop.
Simulation gives us a transferable dexterous skill; the real world gives us true physics and touch. Together, they make data collection dramatically easier.
Grippers are simple: few DoF, robust, with a clean interface like grip/release. Dexterous hands are nothing like that. Now imagine a future where we have a dexterous manipulation skill library as the interface - easy to use, easy to transfer, and perfect for data collection. Imagine how fast that ecosystem could grow.
And as the first step toward that future, we are open-sourcing the entire system: https://t.co/jGpCZJXiUf
We challenged Antigravity to solve the Inverted Pendulum on a custom mechanical system it had never seen before.
Antigravity analyzed hardware specs, coded the control algorithm, and fine-tuned parameters based on performance plots.
See it in action. 👇
Dexterous manipulation is moving fast, and we are excited to be part of that momentum!
Meet the MANUS team together with @AltBionics at the @HumanoidsSummit in Silicon Valley on December 11th and 12th.
Come say hi and see what we have been building. #humanoidssummit
How do we make dexterous hands handle both power and precision tasks with ease? 🫳👌🫰
We introduce Power to Precision (💪➡️🎯), our new paper that optimizes both control and fingertip geometry to unlock robust manipulation from power grasp to fine-grained manipulations.
With simplified finger motions and augmented fingertips, the hand can perform diverse motions from pinching a nut🔩 to handling a pan🍳. Check the demos below🎥.
Huge milestone achieved!📣
World's first mass delivery of humanoid robots has completed! Hundreds of UBTECH #WalkerS2 have been delivered to our partners. 🤖
The future of #industrial automation is here.
March forward to transformation! 🚀
#HumanoidRobots#massproduction#AI
Building a cup tower with a tendon-driven robotic hand ☕🤖
Stable grasping, precise placement, controlled release — three layers up.
What else do you want to see our hand try next?
Drop your ideas in the comments 👇
🧩 CAD: https://t.co/kC213FwuZA
💻 GitHub: https://t.co/H0UhUwB154
🔗 Shop: https://t.co/QeohUpAnUB
#robotics #AI #opensource #dexteroushand #manipulation #embodiedAI
Our SIMA 2 research offers a strong path towards applications in robotics and another step towards AGI in the real world. Find out more → https://t.co/wJKp45K4RW
@wenlong_huang@JunyaoShi The question is not “world models vs. policies” but rather “what level of simulation abstraction/granularity we need for which task?”
Great point on the classic model learning vs policy learning debate! Playing tennis is certainly the case where policy is much easier than the model, but imo the question is largely task/horizon/dynamics dependent. In any case, spatial intelligence underpins both, and a model could be potentially useful for many other stuff both inside robotics (planning, eval, interpretability) and outside (content creation, vr, education, etc)
@drfeifei For Google, the "world" is a latent dynamics interactive video generation model for training agents. For NVIDIA, the "world" is a synthetic data pipeline for training robots. For World Labs, the "world" is a generative, explicit 3D geometric asset for human and AI interaction.
AI’s next frontier is Spatial Intelligence, a technology that will turn seeing into reasoning, perception into action, and imagination into creation. But what is it? Why does it matter? How do we build it? And how can we use it?
Today, I want to share with you my thoughts on building and using world models to unlock spatial intelligence in this essay below. 1/n