Hyunjun Lee @ ICML 2026

Undergrad Intern @ CLVR | ECE & Math @ SNU

[Hyunjun Lee @ ICML 2026 @Hyunjun_Lee1013]

🔥 Excited to share my first work from CLVR Lab!

We introduce N2M a module that guides the robot to the suitable pose for executing the manipulation policy. Check out this post!

[Kaixin Chai @Kaixin_Chai]

🚀 Introducing N2M!

In mobile manipulation, the performance of a manipulation policy is very sensitive to the robot’s initial pose. N2M guides the robot to a suitable pose for executing the manipulation policy.

N2M comes with 5 key features - Check them out in the posts below!

[Hyunjun Lee @ ICML 2026 @Hyunjun_Lee1013]

Presenting my paper at ICML today! 🚀 Stop by poster #513 between 10:30 - 12:15 to chat about our work. Would love to connect and hear your thoughts!

[Hyunjun Lee @ ICML 2026 @Hyunjun_Lee1013]

🔥 Excited to share my first work from CLVR Lab!

We introduce N2M a module that guides the robot to the suitable pose for executing the manipulation policy. Check out this post!

[Sookwan Han @SookwanH]

Temporal Cortex (TC) shouted out by Jensen Huang! @nvidia @NVIDIAAI

TC is the best bio-inspired memory for AI agents: one tht doesnt rot.

For hackers who want long-running AI agent memory, TC is a MUST.

Sign up here: https://t.co/Og9eBm8lH0

@2_9sigma
@im_sihun
@ksunw0209
@nurijunsu

[Lucy Shi @lucy_x_shi]

1/ We just released π0.7 — a steerable generalist robot model with emergent capabilities.
I want to share a bit of the backstory, because π0.7 taught me something surprising about where robot learning is heading. A thread on bittersweet lessons 🧵

[Tairan He @TairanHe99]

Zero teleoperation. Zero real-world data. ➔
Autonomous humanoid loco-manipulation in reality.

Introducing VIRAL: Visual Sim-to-Real at Scale.

We achieved 54 autonomous cycles (walk, stand, place, pick, turn) using a simple recipe:
1. RL
2. Simulation
3. GPUs

Website: https://t.co/tmkKpWsqqW
Arxiv: https://t.co/PKIfxToUGw

Deep dive with me: 🧵

[Yujia Zheng @YujiaZheng9]

🧠 What if large models could read each other’s minds?

Our new paper (#neurips2025 spotlight), “Thought Communication in Multiagent Collaboration”, explores how large model agents can share latent thoughts, not just messages.

📷https://t.co/IzX1Hy7K0K (CMU × Meta AI × MBZUAI)

Imagine teams of agents that don’t just talk, but directly read each other’s minds during collaboration. That’s the essence of Thought Communication, which goes beyond the fundamental limits of natural language, or any observed modalities.

🧩 Theoretically, we prove that in a general nonparametric setting, both shared and private latent thoughts can be identified from model states under a sparsity regularization.

Our theory ensures these recovered representations reflect the true internal process of agent reasoning, and that the causal structure between agents and their thoughts can be reliably recovered.

⚙️ Practically, we introduce ThoughtComm, a general framework for latent thought communication. Guided by the theory, we implement a sparsity-regularized autoencoder to extract thoughts from model states and infer which are shared or private.

This lets agents not only know what others are thinking, but also which thoughts they mutually hold or keep private — a step toward real collective intelligence.

Across diverse models, communication beyond language directly enhances coordination, reasoning, and collaboration among LLM agents.

🔮 In line with recent studies, we believe this work further highlights the importance of the hidden world underlying foundation models, where understanding thought, not just observational behavior, becomes central to intelligence.

#MultiAgent #LLMs #Causality #AI #ML

Joint work with @zhuokaiz @ Zijian Li @xie_yaqi @ Mingze Gao @LizhuZhang @kunkzhang

[Wenli Xiao @_wenlixiao]

What if robots could improve themselves by learning from their own failures in the real-world?

Introducing 𝗣𝗟𝗗 (𝗣𝗿𝗼𝗯𝗲, 𝗟𝗲𝗮𝗿𝗻, 𝗗𝗶𝘀𝘁𝗶𝗹𝗹) — a recipe that enables Vision-Language-Action (VLA) models to self-improve for high-precision manipulation tasks.

PLD couples real-world residual reinforcement learning with standard supervised fine-tuning — letting robots discover, recover, and distill their own data flywheel.

Quick 🧵

[Yanjiang Guo @Yanjiang_Guo]

Rollouts in the real world are slow and expensive. What if we could rollout trajectories entirely inside a world model (WM)?

Introducing 🚀Ctrl-World🚀, a generative manipulation WM that can interact with advanced VLA policy in imagination. 🧵1/6

[Jesse Zhang @Jesse_Y_Zhang]

How can we help *any* image-input policy generalize better?

👉 Meet PEEK 🤖 — a framework that uses VLMs to decide *where* to look and *what* to do, so downstream policies — from ACT, 3D-DA, or even π₀ — generalize more effectively!

🧵

[Jeremy Siburian @jeremy_dasa]

I presented our work "Grounded Vision-Language Interpreter for Integrated Task and Motion Planning" as a poster at the SAFE-ROL Workshop in #CoRL2025!

Thanks to everyone who stopped by and checked out our bimanual cooking demo video!

If you're around CoRL, would love to chat!

[Minho Heo @MinhoHeo]

Come see what robot learning can do for surgical automation!
We’re excited to host the first Workshop on Automating Robotic Surgery with an amazing lineup of speakers.

🗓️ Sept. 27 09:30AM - 12:30PM
📍 Floor 3F, Room E7
🌐 https://t.co/Lp10McSvb2

#CoRL2025 #CoRL @corl_conf

[Jaeyeon (Jay) Kim @Jaeyeon_Kim_0]

Announcing Flexible Masked Diffusion Models (FlexMDMs)—a new diffusion language model for flexible-length sequences.

🚨 Solves MDMs' fixed-length issue + retrains any-order sampling
🚨 <1000 GPU-hrs to fine-tune LLaDA-8B into FlexMDM (GSM8K 58→67%, HumanEval-infill: 52→65%)

[Demis Hassabis @demishassabis]

Official results are in - Gemini achieved gold-medal level in the International Mathematical Olympiad! 🏆 An advanced version was able to solve 5 out of 6 problems. Incredible progress - huge congrats to @lmthang and the team! https://t.co/pp9bXF7rVj

[Sookwan Han @SookwanH]

We can synchronize multiple diffusion models for one collaborative generation. But WHY does it work? HOW should we synchronize?

**SyncSDE** explains
- WHY it works: probabilistic framework for synchronizing diffusion models
- WHERE to focus: where synchronization heuristics should be strategically focused, to reduce trial-and-error
- Strong results on diverse tasks

Accepted in CVPR 2025!!
Arxiv: https://t.co/tQi5bLT37G
Website: https://t.co/GgE6sovWzA

[1/n]