Exciting News! Our new paper "3D-MVP" is out! We propose a novel approach for 3D multi-view pretraining using masked autoencoders, leveraging Robotic View Transformer (RVT) to improve generalization for downstream robotics tasks. jasonqsy.github.io/3DMVP/

🚀 Day 0: Warming up for #OpenSourceWeek! We're a tiny team DeepSeek exploring AGI. Starting next week, we'll be open-sourcing 5 repos, sharing our small but sincere progress with full transparency. These humble building blocks in our online service have been documented,

working reproduction of pi0! awesome stuff

Over the last few months, we’ve been thinking about how to learn from “off-domain” data - data from non-robot sources like video or simulation. These data sources are not quite good enough to learn policies (even monolithic VLA models) directly, but they still contain lots of

There is so much potential in moving beyond simple natural language when building robot foundation models. Trajectories are a great way to do this -- and HAMSTER has taken trajectory-conditioned policies to the scale of VLAs! Congrats to the whole HAMSTER team.

⛔ STOP WHAT YOU'RE DOING ⛔ THERE IS A NEW ROBOT IN TOWN ~ LeKiwi 🥝 ~ ... build it yourself to automate daily choirs with AI 🤗 1/🧵 A thraed!

Its extremely clear to me that this sort of approach is the future, especially in relatively structured warehouse and logistics environments. When I was doing industrial manipulation, it was common for setup to take multiple days. Integration with existing systems and workflows

It is exciting to see another Hierarchical VLA models in the same month! 🎉 Hi Robot from Physical Intelligence uses language, Helix from Figure uses latent embedding, our HAMSTER from @NVIDIA uses 2d paths. We’re all inspired by Thinking, Fast and Slow—combining fast, intuitive (System

Meet our first general-purpose robot at Dexmate dexmate.ai/vega Adjustable height from 0.66m to 2.2m: compact enough for an SUV, tall enough to reach those impossible high shelves. Powerful dual arms (15lbs payload each) and omni-directional mobility for ultimate

Thrilled to announce GR00T N1, our open foundation model for generalist humanoid robots! GR00T N1 adopts a dual-system design, leverages the entire data pyramid for model training, and supports various robot embodiments. GR00T N1 embodies years of fundamental research, spanning

Meet SO-101, next-gen robot arm for all, by Hugging Face 🤗 Enables smooth takeover to boost AI capabilities, faster assembly (20mn), same affordable price ($100 per arm) 🤯 Get yours today! Links in thread below 👇

🎊 That's a wrap on #ICLR2025. Shout out to all the amazing research in #robotics, machine vision, and more. Missed it? Check out our #NVIDIAResearch paper on hierarchical action models for open-world robot manipulation. 📄 nvda.ws/42OZKcC

Was just getting warmed up

On my way to Atlanta to present ZeroMimic: Distilling Robotic Manipulation Skills from Web Videos at IEEE ICRA! Stay tuned for an in-depth post about how ZeroMimic distills zero-shot policies from web human videos. 🌐 Project site: zeromimic.github.io

Introducing 𝐃𝐫𝐞𝐚𝐦𝐆𝐞𝐧! We got humanoid robots to perform totally new 𝑣𝑒𝑟𝑏𝑠 in new environments through video world models. We believe video world models will solve the data problem in robotics. Bringing the paradigm of scaling human hours to GPU hours. Quick 🧵

Imitation learning has a data scarcity problem. Introducing EgoDex from Apple, the largest and most diverse dataset of dexterous human manipulation to date — 829 hours of egocentric video + paired 3D hand poses across 194 tasks. Now on arxiv: arxiv.org/abs/2505.11709 (1/4)