The new "GPT4-o create image" makes the foundation model figure in my brain come true.

3DGSIM: Learning 3D-Gaussian Simulators from RGB Videos Contributions: • Training on RGB video: Avoiding access to meshes or depth cameras, 3DGSim learns particle interactions directly from multi-view RGB images by representing scenes as 3D Gaussian point clouds. • Point

Kids in 2025: writing Tom & Jerry instead of watching them. TTT + diffusion transformer is wild!

#RSS2025 NaVILA constitutes a successful attempt for VILA to drive real world robotic dogs and humanoid! Fully deployable. Money saving. Fast inference. Check out our project page: navila-bot.github.io Many more amazing things to come!

Google presents How new data permeates LLM knowledge and how to dilute it

[1/8] 🚀 New preprint: Climbing the Ladder of Reasoning: What LLMs Can—and Still Can’t—Solve after SFT? We provide a systematic study on the potential and limitations of Supervised Fine-Tuning (SFT) on math reasoning tasks.

Our #ICLR2025 paper MCR will be presented at Hall 3 + Hall 2B #42 on Apr 24th from 7:00 to 9:30 PM PDT. Won't be able to attend the conference since I'm working on CoRL submission. Please check it out and drop me an email if you are interested!

🚀 Excited to announce our 4th Workshop on Computer Vision in the Wild (CVinW) at #CVPR2025 2025! 🔗 computer-vision-in-the-wild.github.io/cvpr-2025/ ⭐We have invinted a great lineup of speakers: Prof. Kaiming He, Prof. Boqing Gong, Prof. Cordelia Schmid, Prof. Ranjay Krishna, Prof. Saining Xie, Prof.

Robot Learning needs 4D world models! Robot Learning needs 4D world models! Robot Learning needs 4D world models! We introduce TesserAct, a 4D embodied world model that can simulate how agents interact with the 3D world over time! We achieve this by simply extending a

Check out the UCSD AMO Humanoid project! It introduces Adaptive Motion Optimization (AMO) for hyper-dexterous humanoid control, enabling real-time, adaptive movements. This framework combines sim-to-real reinforcement learning with trajectory optimization, enhancing stability

Want to scale robot data with simulation, but don’t know how to get large numbers of realistic, diverse, and task-relevant scenes? Our solution: ➊ Pretrain on broad procedural scene data ➋ Steer generation toward downstream objectives 🌐 steerable-scene-generation.github.io 🧵1/8

We've released the code for LegoGPT. This autoregressive model generates physically stable and buildable designs from text prompts, by integrating physics laws and assembly constraints into LLM training and inference. This work is led by PhD students Ava Pun, Kangle Deng,

Congratulations Dr. Mu Cai Mu Cai! Mu is my 8th PhD student and first to start in my group at UW–Madison after my move a few years ago. He made a number of important contributions in multimodal models during his PhD, and recently joined Google DeepMind. I will miss you a lot Mu!

"Building Rome with Convex Optimization" has been accepted to #RSS2025! Try XM, our new structure from motion pipeline powered by GPU-accelerated convex semidefinite optimization: github.com/ComputationalR… XM solves large-scale (nonconvex) global bundle adjustment problem via

On my way to ICRA! Our group will be presenting Mobile-TeleVision (below) and WildMA (wildlma.github.io). Looking forward to chatting!

🚀 Struggling with the lack of high-quality data for AI-driven human-object interaction research? We've got you covered! Introducing HUMOTO, a groundbreaking 4D dataset for human-object interaction, developed with a combination of wearable motion capture, SOTA 6D pose

🚨 BREAKING: NVIDIA JUST announced roadmap for physical AI, robotics and national-scale AI factories. Here’s a breakdown of the top important announcements: 🧵👇 1. DeepSeek R1 is now 4x faster, setting the standard for AI in inference and reasoning.

Slides are here if you are interested: drive.google.com/file/d/1xta-H2…

For years, I’ve been tuning parameters for robot designs and controllers on specific tasks. Now we can automate this on dataset-scale. Introducing Co-Design of Soft Gripper with Neural Physics - a soft gripper trained in simulation to deform while handling load.

How much do language models memorize? "We formally separate memorization into two components: unintended memorization, the information a model contains about a specific dataset, and generalization, the information a model contains about the true data-generation process. When we