Andrei Polubarov

RL, Robotics, Generative Modelling fellow @ dunnolab

[Andrei Polubarov @earo050]

Is it possible to build a multi-domain action model capable of adapting to unseen dynamics?

Check out our new #ICLR2026 paper! We pushed in-context RL scaling further and released Vintix II.

👇👇👇

[Andrei Polubarov @earo050]

@DJiafei I guess VLA need time to become useful. The way they are trained now is inefficient.

[Andrei Polubarov @earo050]

Today at #ICLR2026, we are presenting Vintix II: Decision Pre-Trained Transformer is a Scalable In-Context Reinforcement Learner.

📍 Poster: Pavilion 4, #4516

Happy to chat about in-context RL, robotics, and foundation models for decision making

[Andrei Polubarov @earo050]

We are grateful to @TianheYu, @TokicMichel, @ShengjieWa34067, @mitrma, @Etonwanaa, @teopir, @pengzh97, @zhoubolei and others for building the environments used in this work!

[Andrei Polubarov @earo050]

Is it possible to build a multi-domain action model capable of adapting to unseen dynamics?

Check out our new #ICLR2026 paper! We pushed in-context RL scaling further and released Vintix II.

👇👇👇

[Andrei Polubarov @earo050]

For interesting details and ablations, feel free to read our paper and check out our code!

Project site: https://t.co/cxzY5tPodw
Paper: : https://t.co/7Z6uPixOc1
Code: https://t.co/p9FS6aCKNa
Dataset: https://t.co/kxoeNlqBs3

[Vladislav Kurenkov @vladkurenkov]

We released 87 hours of @LeRobotHF SO 100/101 datasets.

It is a unified, cleaned, and annotated repackage of 598 open-source community datasets (SO100 and SO101), totaling 22,709 episodes, ~9.4M frames, and 563 tasks.

[Vladislav Kurenkov @vladkurenkov]

🚀 Introducing cadrille: a new SOTA model for CAD reconstruction from images, point clouds, and text—all in one framework with the use of RLVR.

Multimodal inputs + RLVR = clean, editable 3D models.

🧵👇

[Denis Tarasov @ML_is_overhyped]

LLMs are amazing because they can learn in context — read, adapt, and act.

Can we do the same for reinforcement learning? That’s the promise of In-Context RL (ICRL).

But existing offline ICRL methods don’t even optimize rewards.

Our new paper shows why RL matters
🧵

[Alexander Nikulin @how_uhh]

🎥 Pre-training VLAs on human videos is tempting — Latent Action Models quickly become an essential part of leading VLAs, like GR00T (@DrJimFan) — but can they effectively handle messy real‐world videos?

In our #ICML paper we give an answer: not yet, at least without some help!

[Ilya Zisman @suessmannn]

🔥 Zero-shot generalization is the dream: adapt instantly, no fine-tuning. It's why LLMs blew up—but it's not just a language modeling thing. It’s happening in RL too.

🚨 @maxsbob21's new paper dives deep into zero-shot RL under shifting dynamics—and why current methods break.

[Alexander Nikulin @how_uhh]

Two papers accepted by #ICML2025!

[Vladislav Kurenkov @vladkurenkov]

Can In-Context RL scale across multiple domains? Our preliminary results suggest it can.

Vintix: Action Model via In-Context Reinforcement Learning -- https://t.co/NMVu2b08TJ