Takeshi ✈️ CoRL 26

Visiting SF -11/8 →Austin -11/10 →Taipei -11/13
Physical AI, Founder from Japan, UX Designer/Researcher

Always be launching!
・2026年中にサンフランシスコで起業する（10月渡米）

[Takeshi ✈️ CoRL 26 @Ken64078764571]

・Industrial NEXT
Robot work cells that can be moved to a new product line. Taught with simulation and demos, aimed at electronics and auto suppliers.
https://t.co/U3KrfO8gtZ

・Acumino
Robot AI for dexterous factory tasks. Hardware agnostic, taught by workers, and says it can deploy in days.
https://t.co/gKd7OanRzP

・Nomadic AI
Finds failures and edge cases in robot and AV video, then turns them into training data.
https://t.co/40lhI2g2LT

[Takeshi ✈️ CoRL 26 @Ken64078764571]

Joined the Mitsubishi Electric Innovation Day at Plug and Play.

Met some robot deployment startups.
One said the same task works across robots and sites. What breaks is when the task changes.

[Takeshi ✈️ CoRL 26 @Ken64078764571]

https://t.co/eH7eLc2ZNa

[Takeshi ✈️ CoRL 26 @Ken64078764571]

Night Hack at Founders Inc!　
Building with a team under time pressure is the best.

[Takeshi ✈️ CoRL 26 @Ken64078764571]

@MissionRobotBay @Scav @EmbodiedCoX https://t.co/GKU3BCGIt4

[Takeshi ✈️ CoRL 26 @Ken64078764571]

Attended the panel event of the deployment moat of Physical AI

3 things I learned tonight:

・hardware breaks before the model does. Sensors, tactile sensors and hands came up again and again.
・the second deployment of the same task is much faster.
・what scales is a task that repeats. Pick one vertical and one bounded job. Data centers were the example: same task, similar sites.

[Takeshi ✈️ CoRL 26 @Ken64078764571]

@rikiya_y 上手い人・普通の人に加え、（あえて）下手な人のデータも学習効果が高いのかもしれませんね！

データ収集、まだまだ誰も正解を知らない領域のようで、モデルも進化するし日進月歩の世界で面白いです😊

[Takeshi ✈️ CoRL 26 @Ken64078764571]

昨日はヒューマノイド用のデータ収集周りの研究ミートアップに参加するため、Redwood Cityへ
（会場は超閑静な住宅街の一室で新鮮でした！）

・人間の実演から、ロボットのDexterity（器用さ）とAgility（身のこなし）をスケール��せる研究
・ロボット学習に必要なデータが足りない課題に対して、人間の実演を訓練データに変換する
（箱の運搬、ハサミの操���など）

手の動きをセンサーで取得→強化学習という流れで、少ないデータをいかに質をあげるか、それをヒューマノイドの技能に変えていくか、が大事そう

[Takeshi ✈️ CoRL 26 @Ken64078764571]

https://t.co/IxBhufocsb

[Takeshi ✈️ CoRL 26 @Ken64078764571]

At the Voxel51 and Nebius workshop in SF today, here is what I heard about physical AI data:

・New robot, old data doesn’t carry over. Even two units of one model can differ.
・Bought data can’t be checked before you commit. One buyer reportedly lost millions to bad camera calibration.
・Country and culture matter too. Japanese data, for example, is seen as valuable: signs, left-side driving, local ways of interacting.

If your team collects robot data, I’d love to hear what has been hardest in practice.

Feel free to DM me.

[Takeshi ✈️ CoRL 26 @Ken64078764571]

@RoboticsCtr @MongoDB @viamrobotics Hi! I heard about Viam at the Long Horizon event and got very interested.

I know it looks full, but I'd love to join tomorrow's hackathon if any spot opens up, even on the waitlist or as an observer.

I code in Python and work with LeRobot and an SO-101.

[Takeshi ✈️ CoRL 26 @Ken64078764571]

@AnneliesGamble Hi Annelies, I'd love a seat.

I'm Takeshi, a founder from Osaka, now in SF, building a tool that checks whether a robot's data still works when the manufacturing line or robot changes.

Curious what breaks first in real deployments.

[Takeshi ✈️ CoRL 26 @Ken64078764571]

https://t.co/ZMbgD4RJVy
来週もあるみたいです。Human-Centered Automationというタイトルからして面白そう

[Takeshi ✈️ CoRL 26 @Ken64078764571]

公開レクチャー（Physical AI: Giving AI a Robot Body）を聴講するため、UC Berkeleyへ

・ロボットは、作業をこなせるかで評価されがちだが、（特に家庭ロボットは）人��の関わりも体験に影響してくる
・ロボットが人間の動きに対応したり、人間同士の関わり（家族の各メンバーに対する振る舞い）を見て動きを変える研究も

データという面では、タスクをこなすための訓練データだけでなく、人との関わりを判断するためのデータ・動き方のデザインも大事になりそう

[Takeshi ✈️ CoRL 26 @Ken64078764571]

Hi! Just moved to SF from Japan/Osaka.

I’m building a data collection platform in the Physical AI space, helping robotics teams check if their data carries over to a new robot.

Would love a coffee with
・data/ML leads at Series A+ robotics cos
・teleop/egocentric/annotation sellers
・robot data QA folks

Who should I meet? Intros welcome!

[Takeshi ✈️ CoRL 26 @Ken64078764571]

ちなみにSFから乗ったCalTrainはほんとにでかかった。新車らしく車内も超快適！