Albina Klepach

🤩Excited to share our #AAAI2026 Oral on #EmbodiedAI: How can robots learn from noisy, unlabeled videos—like those on the internet—without action labels? Our new work, Object-Centric Latent Action Learning, tackles exactly this. Thread ↓

For those who haven’t seen our work on latent action models, here is a thread

Thanks to @RealAAAI for the conference. So many interesting new connections. Hope that now more people are aware of the beauty of latent actions for robotics 🤖 #AAAI2026 #robots #ai #aaai #Singapour

Don't forget to visit my oral or/and poster today on scalable imitation learning via object-centric pretraining. Oral: Garnet 218 @ 9:50 am Poster: Hall 4 Also, feel free to reach out if you want to chat about continual, imitation and/or reinforcement learning while at AAAI!

@AmitLeViAI Come across my paper on scalable imitation learning

Drop a link to your #AAAI2026 paper in the comments if you’re coming this week👇 #AAAI starts soon, and I’m heading there tomorrow(but I will actually arrive on the day after🥲) ✈️ I’d love to meet fellow researchers, industry professionals, and investors who are passionate about AI, to exchange ideas, get feedback, or explore potential collaborations. My ongoing research focuses on LLMs, including (but not limited to): •Safety: Evaluation, Alignment •Interpretability •Unlearning •Continual learning •Pre-training •Model optimization and compression •Diffusion LMs •AI Agents Security If you’re attending AAAI, feel free to message me 😊 See you at the conference!

🙏Thanks to my great co-authors! @how_uhh @suessmannn @vladkurenkov And thanks to all for inspiring research on latent actions @jparkerholder @_rockt @svlevine @SeonghyeonYe @jang_yoel @furongh and on slot-attention @FrancescoLocat8 @tkipf @Dazitu_616

9/9 If you're at AAAI or interested in world models, imitation learning, or object-centric RL, check out our paper: arxiv.org/abs/2502.09680 Oral session: Jan 22, 9:30-10:00 Garnet 218 Poster session: Jan 22, Hall 4 Code (coming soon): github.com/dunnolab/objec…

🤩Excited to share our #AAAI2026 Oral on #EmbodiedAI: How can robots learn from noisy, unlabeled videos—like those on the internet—without action labels? Our new work, Object-Centric Latent Action Learning, tackles exactly this. Thread ↓

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.

I’m asking for help. I was meant to start my PhD with @_rockt and @robertarail at UCL, but my UK background check was refused. My appeal seems unlikely to succeed, so I’m urgently searching for any PhD or research positions in academia or industry. Any help is appreciated.

🚀 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. 🧵👇

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 🧵

🎥 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!

🔥 Φ-Module 🔥 We present Φ-Module - an easy way to compute partial charges and electrostatic energy in a self-supervised way for any neural network interatomic potential. Learn more below!