We figured out how to train VLAs with diffusion outputs much faster (7.5x faster), inheriting better language following from the VLM, and leading to better results. The key: protect the VLM backbone during training with knowledge insulation. Let’s talk about what we learned👇

Can we collect robot dexterous hand data directly with human hand? Introducing DexUMI: 0 teleoperation and 0 re-targeting dexterous hand data collection system → autonomously complete precise, long-horizon and contact-rich tasks Project Page: dex-umi.github.io

Best platform for loco-manipulation research. RobotEra XHAND connected to Unitree G1 for teleoperation. #HumanoidRobot #dextroushand #xhand #unitreeG1 #robotera #Unitree

Significant progress in AI and Robotics this week. So, I summarized everything from Google DeepMind, Anthropic, Figure, DeepSeek, Sakana AI, Perplexity, ETH Zurich, Unitree, Physical Intelligence, and more. Here's everything you need to know and how to make sense out of it:

This looks incredible!

In LLM land, a slow model is annoying. In robotics, a slow model can be disastrous! Visible pauses at best, dangerously jerky motions at worst. But large VLAs are slow by nature. What can we do about this? An in-depth 🧵:

Cool!

We’re releasing the RoboArena today!🤖🦾 Fair & scalable evaluation is a major bottleneck for research on generalist policies. We’re hoping that RoboArena can help! We provide data, model code & sim evals for debugging! Submit your policies today and join the leaderboard! :) 🧵

History is a very important topic in robotics, but always be ignored. This work shows a solution for incorporating history. While I think incorporating history through reasoning is also a good idea, like what we do in OneTwoVLA.

Here's a F.02 in my home, using Helix to do my laundry

“Doing research in AI is about understanding intelligence itself,” says Wen Chuan Chuan Wen (China), alumnus of the Institute for Interdisciplinary Information Sciences, who explores embodied intelligence, learning how robots reason causality & spatial relationships.

Excited to share that HuB has been accepted for an Oral Presentation at CoRL 2025! See you in Seoul this September.

Introducing DINOv3 🦕🦕🦕 A SotA-enabling vision foundation model, trained with pure self-supervised learning (SSL) at scale. High quality dense features, combining unprecedented semantic and geometric scene understanding. Three reasons why this matters…

The motion looks super human-like