Junho Cho (Jacey)

Bridging AI research & strategy @RLWRLD_ai | ex-Samsung | PhD SNU, KAIST alum

[Junho Cho (Jacey) @junh0ch0]

If you're at #ICML2026, feel free to dm me and meet & talk about #VLA #RFM, humanoid dexterous manipulation, human data, and real-world deployment.

We also host Robotics Night at the @RLWRLD_ai office (near COEX) evening 7/8. See you there!

[Junho Cho (Jacey) @junh0ch0]

@ShenyuanGao Looking forward to see this work!

[Junho Cho (Jacey) @junh0ch0]

@jadenvclark Great work! We better have more sensory data other than vision!

[RLWRLD @RLWRLD_ai]

Robotics Night is coming to Seoul 🇰🇷🤖

Presenting at #ICML2026? Deep into robotics AI? Come spend an evening with fellow researchers at RLWRLD's Seoul research lab on July 8: live RLDX-1 demos on real robot hands, drinks, and casual talk with the people building the model. 👇🏻

🤝 Who you'll meet: ICML authors and presenters across VLA, world models, manipulation, and dexterity, mingling with the RLWRLD researchers who built RLDX-1
🤖 Live RLDX-1 demos on real robot hands, our dexterity-first foundation model for true five-finger dexterity
📍 July 8 (Wed), 7:30 PM · RLWRLD Lab, Gangnam, Seoul (2 blocks from COEX where ICML2026 is)
🎟️ Invite-only, seats are limited

Want in? Find someone from RLWRLD at ICML, or DM @junh0ch0 (Jacey) and we'll get you on the list. 👀

Preview of what runs live that night 👉 https://t.co/eLFhxjiW2i
Here's our story 👉 https://t.co/VgOSjan1Iv
Check out the RLDX-1 tech blog 👉 https://t.co/eB6ZK1thm4

#RLDX #RLWRLD #ICML2026 #PhysicalAI #Robotics #DexterousManipulation #RoboticsFoundationModel #VLA #WorldModel

[Andrew Kang @Rewkang]

NVIDIA went public at a $625M Valuation

Now companies like OpenAI, Anthropic, SpaceX do not go public until they are $1 Trillion +

Public Markets need Venture Capital Access

[LeRobot @LeRobotHF]

Check out this fantastic video created by @_jong_hyun_park for his channel and Korean audience! It offers a great behind-the-scenes look at what we do at the LeRobot team. We had a wonderful time doing the interview, and the video is packed with interesting insights.

https://t.co/8Igqpr7b3n

[Junho Cho (Jacey) @junh0ch0]

Introducing RLDX-1

[RLWRLD @RLWRLD_ai]

(1/12) Hi, we are #RLWRLD (ReaL WoRLD).

RLDX-1 is live. Dexterity is intelligence, and it lives in #RLDX (RealDex).

A dexterity-first foundation model from for robot hands that builds muscle memory through motion, history, and contact.

- Across 10 dexterous real-world tasks on the ALLEX humanoid and DROID setup, RLDX-1 ~2× outperforms π₀.₅ and GR00T N1.6.
- SOTA on 8 simulation benchmarks in LIBERO, SIMPLER, and RoboCasa.

Everything ships today:
training and inference code, the pre-trained model, mid-trained checkpoints, and a fine-tuned checkpoint for every reported benchmark.
LoRA recipes for parameter-efficient adaptation to your robot. Supports LeRobot v2.1 datasets. A one-line CLI to toggle motion / memory / physics modules per embodiment.
Full technical report with ablations.

Not just our cool demos. We document the architecture, data, and design decisions behind every result. Bring your own robot. Train and deploy RLDX-1 on it.

🧵 What's inside RLDX-1, in 12 posts.

🌐 https://t.co/HjoLIOeHDr
📄 https://t.co/JRDceLah7s
💻 https://t.co/IO441l69av
🤗 https://t.co/7W2XztYiSx

[RLWRLD @RLWRLD_ai]

#RLWRLD is at #ICLR2026 in Rio 🇧🇷 and three of our papers are on the floor this week. Drop by the posters.

And also #RLDX-1, our dexterity-first foundation model for robot hands, drops in the coming weeks. Join the launch list and we'll ping you the day it drops 👉 https://t.co/u1Qe48LSmn

HAMLET: Switch your VLA into a History-Aware Policy — @myungkyukoo et al.
A memory module with moment tokens that lets VLAs encode and integrate past observations.
📍 Apr 24, AM–1:00 PM · Pavilion 3 · P3-#1317
🪧 https://t.co/CoJiu7JLkE
🌐 https://t.co/QOWUD3U9LM
📄 https://t.co/J2F3KydbgF
👀 Spoiler alert: this memory module is one of the building blocks inside RLDX — more on that soon.

MG-Select: Verifier-Free Test-Time Sampling for VLAs — @gloryhyeok24 et al.
Training-free test-time scaling via KL divergence as a confidence signal for action selection.
📍 Apr 23, 3:15–5:45 PM · Pavilion 3 · P3-#1303
🪧 https://t.co/1z6ObZej1O
🌐 https://t.co/h4utqPSJiH
📄 https://t.co/ZfjSjMWCfa

Target-Aware Video Diffusion Models — @taeksu98 et al.
Video generation where the actor accurately interacts with a target specified by a segmentation mask.
📍 Apr 24, AM–1:00 PM · Pavilion 4 · P4-#3202
🪧 https://t.co/CbYr9NbxRA
🌐 https://t.co/fYVrvowIbN
📄 https://t.co/uBtPOmQEWQ

#VLA #RFM #PhysicalAI #Robotics #RobotLearning #Manipulation #VideoGeneration

[RLWRLD @RLWRLD_ai]

Three identical boxes. A mouse is placed into one. A moment later, a go signal, and the robot has to pick.
Without memory, the policy forgets which box.
RLDX-1's Memory Module is built on HAMLET (#ICLR2026 in Rio 🇧🇷), integrated into the full architecture. Open-sourced in two weeks. Stay tuned. 👉 https://t.co/u1Qe48LSmn

#RLDX #RLWRLD #PhysicalAI #Robotics #Dexterity #FoundationModels #Automation #Manufacturing #VLA

[Junho Cho (Jacey) @junh0ch0]

Coming in few weeks!

[RLWRLD @RLWRLD_ai]

The human hand is the most sophisticated tool ever created.

What if AI could finally understand it?

At RLWRLD, we call it Dexterity Intelligence — not just robots that move, but AI that thinks through its hands.

Our first proprietary model, RLDX-1, is coming soon. We've reserved a spot for those who want to experience it first.

👉 https://t.co/u1Qe48LSmn

[RLWRLD @RLWRLD_ai]

6 papers from RLWRLD accepted to CVPR 2026! 🎉 Congrats to all the authors and collaborators, led by Professor Joo (@jhugestar) and Professor Cho (@LabyrinthMaker) 👇🏻

📚 DextER: Language-driven Dexterous Grasp Generation with Embodied Reasoning — contact-based embodied reasoning that predicts finger-link contacts on object surfaces for language-driven dexterous grasping
Junha Lee, Eunha Park (@eunha724), Minsu Cho
🔗 https://t.co/EMe7tRYnw2
🔗 https://t.co/ff80Tg9K6r

📚 Dexterous World Models — scene-action-conditioned video diffusion model to simulate embodied dexterous actions in a given static 3D scene
Byungjun Kim*(@byungjun__kim), Taeksoo Kim*(@taeksu98), Junyoung Lee(@junc0ng), Hanbyul Joo
🔗 https://t.co/Hnoz4EdpRU

📚 Affostruction: 3D Affordance Grounding with Generative Reconstruction — generative reconstruction to complete occluded regions and ground affordances on full 3D shapes
Chunghyun Park, Seunghyeon Lee (@llishyun), Minsu Cho
🔗 https://t.co/TqAntMRbdd
🔗 https://t.co/bbgkizuxVq

📚 Planning in 8 Tokens: A Compact Discrete Tokenizer for Latent World Model — CompACT compresses each observation into just 8 discrete tokens, enabling orders-of-magnitude faster planning in latent world models
Dongwon Kim (@dngwnkm), Gawon Seo, Jinsung Lee, Minsu Cho, Suha Kwak
🔗 https://t.co/qB7HohHYq1

📚 MoGaF: Space-Time Forecasting of Dynamic Scenes with Motion-aware Gaussian Grouping — long-term stable scene forecasting via motion-aware Gaussian grouping
Junmyeong Lee* (@ijunmye02373079), Hoseung Choi*, Minsu Cho
🔗 https://t.co/Io0Mv7lJbO

📚 Improving Text-to-Image Generation with Intrinsic Self-Confidence Rewards — post-training T2I generators with the model's own self-confidence as reward, improving compositionality and text-image alignment without external reward models
Seungwook Kim (@1ndependentgrad), Minsu Cho
🔗 https://t.co/EDwsxQCH3V

[Junho Cho (Jacey) @junh0ch0]

@ShenyuanGao Great work! Is DreamDojo-HV (44k hr) dataset is going to be public as well?

[Junho Cho (Jacey) @junh0ch0]

Generating not only images but other modaility as 3d lidar? What if it also can generate force sensing in contact-rich manipulation?

[Google DeepMind @GoogleDeepMind]

@Waymo The model transfers Genie 3’s vast world knowledge into precise camera and 3D lidar data unique to Waymo’s hardware.

Engineers can prompt “what if” scenarios – like extreme weather or reckless drivers – to stress-test the system.

[Junho Cho (Jacey) @junh0ch0]

@NVIDIAAI @DrJimFan @jang_yoel @moo_jin_kim @kvablack Video World Model is a meaningful step forward for representation and planning. My sense is deployable robot AI will also need fast closed-loop execution, non-visual sensor fusion, and multi-layer async architectures. Excited to see where this goes.

[Junho Cho (Jacey) @junh0ch0]

Really interesting works from @nvidiaai — DreamZero (@drjimfan, @jang_yoel) and Cosmos Policy (@moo_jin_kim) make a compelling case for Video World Models over VLAs. I'm largely on board. Just wanted to share a few thoughts on where the gaps might still be.

[Joel Jang @jang_yoel]

Robotics right now feels like peak entropy. Everyone has a different bet on what will work, and they're all confident, which is why doing robotics research right now is so fun.

I wrote an essay on the question that's been driving our work from DreamGen → DreamZero → what’s next

My bet: human experience is the only data source that scales, world models are the right paradigm, and humanoids have the edge.

https://t.co/H1Pvr4R34l

[Junho Cho (Jacey) @junh0ch0]

@NVIDIAAI @DrJimFan @jang_yoel @moo_jin_kim @kvablack On data: video can scale via smart glasses, but tactile/force data has no scalable human collection method. Motion capture (hardware-agnostic) and teleoperation (hardware-specific) are complementary, not competing like web text vs RLHF data in LLMs.