
Head of AI and Robotics at @airoa_org.
日本 東京
Joined September 2011
- Tweets17K
- Following697
- Followers1.6K
- Likes10.5K
Thank you very much for participating in the UMI Arena workshop and giving an insightful keynote presentation @SongShuran !
It was awesome to see so many creative UMI designs at the UMI Arena @ #CORL2026 ! What an incredible and vibrant community!!!
Kei Ota retweeted
A memorable moment from #CoRL2026 🤖
At our workshop, “The UMI Arena: From Handheld Devices to Real-World Policies,” we were honored to welcome Prof. Shuran Song , the creator of UMI, as an invited speaker.
The UMI Arena was created to bring together knowledge from across the UMI ecosystem — hardware, software, data, models, and real-robot deployment — and to explore how different UMI systems can be compared and evaluated on a more common and fair basis.
We were delighted that Prof. Song shared the vision behind the workshop. After the event, she told us how wonderful it was to see the community “thriving, growing, and making such exciting progress.”
We also had the chance to show her AIRoA’s Franka demo and take a photo together with AIRoA President Tetsuya Ogata and members of our development team.
As the community that has grown around UMI continues to expand, we’re very happy to have helped create a place where researchers and developers could come together, compare ideas, and learn from one another.
Thank you, @SongShuran , for the inspiring talk and for bringing such great energy to the workshop!
#AIRoA #CoRL2026 #Robotics #RobotLearning #PhysicalAI #UMI
Kei Ota retweeted
Today at #CoRL2026 — The UMI Arena: From Handheld Devices to Real-World Policies 🤖
After months of preparation, we’re excited to bring the UMI Arena to Austin today.
The workshop brings together researchers working across data collection, robot learning, and real-world deployment through three tracks:
🦾 Policy Competition — evaluating policies on shared real robots and common tasks
🖐️ Device Exhibition — hands-on trials of UMI-style data-collection devices
📄 Contributed Papers — new research on data collection, cross-embodiment deployment, VLA models, tactile sensing, benchmarking, and more
The program includes 30 accepted papers, 13 devices, 6 spotlight talks, live robot demonstrations, and invited talks.
🏆 Awards will also be presented for Best Policy, Best Device, and Best Paper.
📅 Today, Nov. 12 | 14:00–18:00
📍 JW 303–304, JW Marriott Austin
If you’re at CoRL, come join us this afternoon for hands-on demos, real-robot evaluation, and discussions around the future of UMI-based robot learning.
Workshop details:
umi-arena.airoa.io/
#AIRoA #CoRL2026 #RobotLearning #Robotics #PhysicalAI #UMI #Workshop #Awards
We are excited to announce that AIRoA’s workshop proposal, “The UMI Arena,” has been accepted at CoRL 2026!
At the UMI Arena, participating teams will compete by deploying their robot control policies on shared tasks in a common real-robot environment.
To support model development, a subset of real-world robot manipulation data collected using the YUBI data-collection device will be provided exclusively to participating teams, together with a reference policy model and training code. We plan to expand the data available to participants in stages to approximately 20,000 hours by September 2026.
The workshop will feature three tracks:
・Policy Competition
・Data-Collection Device Exhibition
・Paper and Poster Session Through
these tracks, we aim to build a shared real-robot evaluation platform and advance the standardization of evaluation methods.
Invited speakers are expected to include Shuran Song, Andy Zeng, Hao-Shu Fang, and Yang Gao.
■ November 12, 2026 | Austin,
Texas Learn more ▼
airoa.org/updates/20260821/6…
Workshop website ▼
umi-arena.airoa.io/
#AIRoA #CoRL2026 #RobotLearning #Robotics #PhysicalAI #YUBI
🚀 AIRoA is at #CoRL2026!
📄 8 posters (3 main, 1 LBR, 4 workshop)
🤖 A-0 live demo @ Demo 7
🏟️ Hosting UMI Arena workshop
📍 Booth #210
🍽️ Dinner tonight: grab a ticket at our booth!
Unfortunately I couldn’t make it this year, but ~10 of our members are there. Go say hi! 🙌
📍 Where to find AIRoA at CoRL 2026
Not sure how to get from our exhibition booth on the 2nd floor to the Demo Area on the 3rd floor?
We made a short video to show you the way 👇
Today, Nov. 10, you can find us at:
🤖 Exhibition Booth #210 — 2F
Meet the AIRoA team, try YUBI, see the ELEY demo, and take a spin on our YUBI-powered gacha.
🦾 Demo 7 — 3F
See A-0 / YUBI on Franka in action throughout the day.
And you can find more AIRoA research across CoRL:
✏️AIRoA-0: a robot foundation model trained on 16,151 hours of handheld YUBI data, released with open weights and code.
Led by Yueh-Hua Wu & Kei Ota with the AIRoA team. Booth 210 and Demo 7.
Project: a-0.airoa.io/
✏️S2 (See Less Specify More): goal-preserving local language plus a learned visual evidence budget, so VLAs act only from task-sufficient cues.
Led by Yueh-Hua Wu. Poster 25, Tue, Nov 10 PM.
Project: s2.airoa.io/
✏️Counter Align: counterfactual relabeling that turns expert demos into an instruction-grounded reward for offline RL.
Led by Haru Kondoh and Yueh-Hua Wu, Poster 31, Tue, Nov 10 PM.
Project: counteralign.airoa.io/
✏️YUBI: a finger-aligned handheld gripper for bimanual data collection at scale, with open hardware, software and data.
Led by Take Ohkawa & Jumpei Arima. Poster 267, Wed Nov 11 AM and demo session.
Project: yubi.airoa.io/
✏️YUBI-Real2Sim: 3DGS scene verification that tells when a real-to-sim scene is good enough to rank policies. By Hiroki Nishizawa. Late Breaking Results & UMI Arena Paper track, Tue Nov 10 lunch, board 126.
Paper: lnkd.in/g8CgMB39
✏️YUBI-STAG: automatic contact- and object-grounded language for demonstrations that aligns VLAs with fine-grained instructions.
Led by Masatoshi Tateno and Take Ohkawa. Spotlight talk at The UMI Arena.
Project: lnkd.in/gwtMwhyT
✏️RoboPace: contact-aware retiming that runs action chunks fast in free space and slows down at contact, with no retraining.
Led by Mimo Shirasaka and Kei Ota. UMI Arena Paper track.
Project: lnkd.in/gm8G5ieR
✏️YUBI in the Dark: tactile sensing on a handheld gripper that guides manipulation when vision fails.
Led by Yuma Shin with Asako Kanezaki, Rei Kawakami, and Kei Ota. UMI Arena Paper track.
Paper list: lnkd.in/gHCTQVUX
✏️YUBI Go: an open-source Quest app that records bimanual YUBI demonstrations anywhere. Led by Yuki Noguchi with Yuya Kanai and Kengo Ikeuchi. UMI Arena Device Track.
Code: lnkd.in/gD3DbxS5
🍽️ Networking Dinner — Nov. 10 evening
AIRoA is also hosting a networking dinner tonight. If you’re interested in joining, feel free to send us a DM.
Come find us at the booth, demo, posters, talks — and dinner!
#AIRoA #CoRL2026 #RobotLearning #PhysicalAI #Robotics #VLA #Demo
Kei Ota retweeted
More evidence UMI is the way forward
Introducing A-0: a complete robot learning system built on 16k hours of human demonstrations, with zero hours of robot data collected.
It spans data processing, pre-training, post-training and the inference harness, and it's embodiment-agnostic: the same weights run long-horizon, precise manipulation on four different robots. All the data is recorded with YUBI, our handheld gripper.
▶️ More demos and details: a-0.airoa.io
The paper is out today, and we're releasing the code, weights and dataset: arxiv.org/abs/2611.03815
Introducing A-0: a complete robot learning system built on 16k hours of human demonstrations, with zero hours of robot data collected.
It spans data processing, pre-training, post-training and the inference harness, and it's embodiment-agnostic: the same weights run long-horizon, precise manipulation on four different robots. All the data is recorded with YUBI, our handheld gripper.
▶️ More demos and details: a-0.airoa.io
The paper is out today, and we're releasing the code, weights and dataset: arxiv.org/abs/2611.03815
We're releasing:
📄 Report: arxiv.org/abs/2611.03815
💻 Code: github.com/airoa-org/AIRoA-0
🤗 Weights: huggingface.co/airoa-org/AIR…
📦 Full 16K-hour dataset (coming soon)
Build the gripper, record a task by hand, post-train A-0.
Kei Ota retweeted
AIRoAの研究成果3件と、YUBI(Yielding Universal Bidigital Interface)を用いたデモプロポーザルが #CoRL2026 に採択されました🎉
会期中は、
・研究成果3件の発表
・11/9–10 CoRL公式Demo AreaでのYUBI体験型デモ
・11/12 Workshop「The UMI Arena」主催
・Booth 210でFranka + YUBI / ELEYの実機デモ
を予定しています。
📍Austin, TX
🔗 airoa.org/ja/updates/2026102…
#CoRL2026 #RobotLearning #Robotics #PhysicalAI
Kei Ota retweeted
《日経Robotics》論文には出てこない、YUBIのお話─ハンドヘルド型データ収集デバイスの活用の裏側 xtech.nikkei.com/atcl/nxt/ma…
Kei Ota retweeted
Exciting updates for YUBI at #CoRL2026! 🚀🤖
📄 YUBI Paper Accepted!
Our YUBI paper has been accepted to the CoRL main conference and the Demo track! Please check out our presentation or visit our AIRoA booth to chat with the team.
Project: yubi.airoa.io/
(1/N)
Heading to #IROS2026! 🤖
Excited to present work led by @Yuki_Doradora on non-prehensile manipulation.
ieeexplore.ieee.org/abstract…
I’ll also organize IARL 2026 on industrial robot learning. Join us! 👇
aistc.github.io/IROS2026-wor…
#AIRoA is hiring too! Interested? Come talk to me!
Kei Ota retweeted
YUBI公開後、多くの方に評価いただき感謝しております。皆さまのご意見を考慮し、HWをアップデートしました。詳細はコチラをご覧ください。
github.com/Toyota/yubi-hw/re…
Two papers accepted to #CoRL2026 🎉
・See Less, Specify More: Visual Evidence Budgets for Generalizable VLAs s2.airoa.io/
・CounterAlign: Counterfactual Supervision for Vision-Language-Action Models counteralign.airoa.io/
Grateful to my collaborators at AIRoA. See you in Austin this November!
VLA generalization is hard in part because we ask a robot policy to solve two under-specified problems at once, with far less data than today’s VLMs.
From coarse goals and full images alone, it must infer both what to do now and what in the scene matters for control, then turn that into precise actions from scarce, embodiment-specific robot data.
In See Less, Specify More, we reduce that burden. S2 uses an off-the-shelf VLM for broader reasoning and goal-preserving local guidance, including trajectory-aware execution details, then trains the action policy to act from annotation-free, task-sufficient visual evidence instead of broad scene context.
Across 8 real-robot tasks on TX-G2 and HSR, S2 raises mean subtask success from 54.2% to 79.0% over pi05
🚀 3 papers accepted to CoRL! 🚀
S2 — s2.airoa.io/
YUBI — yubi.airoa.io/
CounterAlign — counteralign.airoa.io/
Huge congrats to @yh_kris @tkhkaeio @arijun37 and everyone at @airoa_org! 🔥
Proud of what the team has built, and excited for what comes next! 🚀
Kei Ota retweeted
データ収集デバイス「YUBI」を活用した大規模なロボットデータ収集のためのオープン・プラットフォームに、参加しませんか?
■「YUBI」の主な特徴
軽量グリッパー設計: 大規模な操作データ収集を容易に
完全オープンソース: ハード(CAD/3Dデータ)もソフトも全公開。ゼロから開発不要
再現性の高い導入ルート: 誰でもすぐに自作・実装が可能
9/2-4のRSJ(金沢)でもブース出展し、YUBIのデモ展示を予定しています。ぜひお立ち寄りください
▼こちらからご視聴いただけます。
youtu.be/o2pTYOngsho?si=2z0n…
Special Thanks : @TOYOTA_FRC
#AIRoA #Robotics #ロボット #オープンソース #エンジニア #RSJ2026













