sean

ai and robots

[sean @seann999]

@liyitengx Door opening

[海外の万国反応記 @all_nations2]

嘘だと言ってくれ
アメリカ人は日本の家は狭いと言いながら芝刈り機動かして庭でBBQしてるはずなんだ

[トヨタ 未来創生センター | Toyota Frontier Research Center @TOYOTA_FRC]

YUBI (Yielding Universal Bidigital Interface), a research-oriented robotic interface developed by the #FrontierResearchCenter, will be showcased at the @airoa_jp booth (#ICRA2026 / Booth 073).
YUBI is designed for intuitive bimanual robot operation and scalable data collection. +

[トヨタ 未来創生センター | Toyota Frontier Research Center @TOYOTA_FRC]

On June 5 at the #ICRA2026 Workshop “Beyond Teleoperation: Learning from Diverse Human and Simulation Data,” AIRoA (@airoa_jp) and the #FrontierResearchCenter will present:
“YUBI: Yielding Universal Bidigital Interface for Scalable Data Curation in Robotic Manipulation.” +

[Jason Shuman @JasonrShuman]

Turns out that Robotics might be the biggest driver of growth for VR Headset sales

[sean @seann999]

@GeneralistAI can GEN-1 draw on it with a marker though, like a nice circle?

[Science Robotics @SciRobotics]

The April 2026 issue of Science #Robotics is out!

This month's cover highlights large behavior models that can outperform single-task policies at complex manipulation tasks like installing a bike rotor. Learn about this research and more: https://t.co/89AzyDYh0O

[Taku🤖🧠 @taku_sid]

フィジカルAIの公募
AIの部分にしか予算ついてない問題

脳だけ賢くしても、モーターが回らなきゃただの置物なんよ🥺
フィジカルAIの「フィジカル」に金出してくれ。センサー、アクチュエータ、筐体、電源。ここをケチったら、日本のロボットは永遠にデモ動画止まり。

[sean @seann999]

@VilleKuosmanen Apply tricks used in TD3? Use 2 target Q networks, use the lower value. Update Q more frequently than pi. Etc.

[Daniel Havir @danielhavir]

Aspire not to build robots. Aspire to build machines. It’s only called a robot when it doesn’t work. The moment it works, it becomes a machine.

[Jen Zhu @jenzhuscott]

If you are starting a robotic / hardware company:

resist humanoid.

resist AI glasses.

[sean @seann999]

@RealJoshuaYang @liyitengx I bought them from Amazon Japan. For Amazon US I might buy these. https://t.co/tKda5qLKx7

[sean @seann999]

80% of AlohaMini with SO107 arms and XLeRobot grippers. Kudos @liyitengx
Awesome that you can build this for less than $1K.
Just teleop for now🎄

[sean @seann999]

@zshobbs @liyitengx I'll probably use this https://t.co/u9KUZZTin1 for cameras. Just L/R wrist cameras for now, maybe add a head camera later on.

[sean @seann999]

@liyitengx Door opening

[sean @seann999]

@liyitengx Bottle opening

[sean @seann999]

@binghao_huang @YunzhuLiYZ @XinyueYolandaZ Is V2 gonna be open-sourced?

[sean @seann999]

Awesome. As a long-time UMI fan myself, it's sometimes surprisingly hard to convince people that this is the way to go. And you need a full-stack team with this conviction to achieve what Sunday did. Hopefully more people will wake up now.

[Tony Zhao @tonyzzhao]

Instead of teleoperation, we train solely on data from our Skill Capture Glove.

The glove is co-designed with Memo's hand, meaning they share the exact same geometry and sensor suite.

If you can do it wearing the glove, Memo can learn it.

[sean @seann999]

While every other robotics company or lab is reproducing a system like ALOHA, Tony's old work, his company is already way past that

[Sunday @sundayrobotics]

November 19

[Simon Kalouche @simonkalouche]

Massive dataset but egocentric video is not going to get us dexterous manipulation policies.

Egocentric data provides high-level semantic scene and task understanding (which frontier VLMs already generally provide). What is needed is fine sub-mm-level finger pose & force data.

[Chris Paxton @chris_j_paxton]

@VectorWang2 I dont disagree lol. I just really hope this thing can touch the floor some day