RT @jyo_pari: What if an LLM could update its own weights?. Meet SEAL🦭: a framework where LLMs generate their own training data (self-edits….

RT @anagh_malik: 📢📢📢 Neural Inverse Rendering from Propagating Light 💡. Our CVPR Oral introduces the first method for multiview neural inve….

RT @ryan_hoque: Imitation learning has a data scarcity problem. Introducing EgoDex from Apple, the largest and most diverse dataset of de….

RT @NicoBohlinger: ⚡️ Can one policy control 1000 different robots? 🤖. We explore Embodiment Scaling Laws: Training on more diverse robot e….

RT @jyo_pari: Llama 4 (@Meta) shows too much SFT limits RL exploration — something we also found in our recent work! A new and superior pre….

Happy to share to be a recipient of the 2025 Apple Scholars in AI/ML PhD Fellowship Programm!. Grateful to @apple for this recognition and the support. Thank you to all my collaborators, colleagues and to supervisor who made this possible!.

RT @NicoBohlinger: ⚡️ Do you think training robot locomotion needs large scale simulation? Think again!. Our new paper shows how to train a….

RT @jyo_pari: Turn a single pre-trained model’s layers into MoE “experts” and reuse them? Finetuning a “router” slightly cuts loss—cool pro….

This was a very fun collaboration with @jyo_pari!.

Pretrained and finetuned checkpoints:. Code: Paper Link:

Results:.- Pretrained on 200k trajectories subset of OXE using just 6 A6000 GPUs for 3 days.- Outperforms OpenVLA and Octo on SIMPLER benchmark .- Strong results on CALVIN ABC (4.01) and SOTA on LIBERO-90 (0.95) with just 20k steps fine-tuning on 4 GPUs in less than 6 hours

Key ideas:.- Noise-conditioned routing lets experts specialize in specific denoising phases.- Expert caching system precomputes and reuses relevant experts for faster inference.- Parameter-free noise injection through noise-conditioned attention to replace AdaLN

How can we make diffusion policies more computationally efficient while scaling up towards generalist policies?. Introducing MoDE: A novel generalist MoE-based Diffusion Policy

RT @hongyi_96: #NeurIPS2024 I'll be in Vancouver from December 8-16, presenting our work "Variational Distillation of Diffusion Policies in….

RT @oier_mees: Tired of labeling your robot data?🤖 .Excited to present NILS, a novel approach that leverages foundation models 🧠 to segment….

RT @Jan_R_Peters: Fantastic gang from @TUDarmstadt @ias_tudarmstadt @GeorgiaChal and offspring at @KITKarlsruhe …

RT @nikos_gkanats: 3D representations are more critical than what we thought for manipulation. Our work 3D Diffuser Actor marries those wit….