RT @stepjamUK: 𝗜'𝘃𝗲 𝗵𝗲𝗮𝗿𝗱 𝘁𝗵𝗶𝘀 𝗮 𝗹𝗼𝘁 𝗿𝗲𝗰𝗲𝗻𝘁𝗹𝘆: "𝗪𝗲 𝘁𝗿𝗮𝗶𝗻𝗲𝗱 𝗼𝘂𝗿 𝗿𝗼𝗯𝗼𝘁 𝗼𝗻 𝗼𝗻𝗲 𝗼𝗯𝗷𝗲𝗰𝘁 𝗮𝗻𝗱 𝗶𝘁 𝗴𝗲𝗻𝗲𝗿𝗮𝗹𝗶𝘀𝗲𝗱 𝘁𝗼 𝗮 𝗻𝗼𝘃𝗲𝗹 𝗼𝗯𝗷𝗲𝗰𝘁 - 𝘁𝗵𝗲𝘀𝗲 𝗻𝗲𝘄 𝗩𝗟𝗔 𝗺𝗼𝗱….

RT @marionlepert: Introducing Masquerade 🎭: We edit in-the-wild videos to look like robot demos, and find that co-training policies with th….

What do you want to know?.

💡6/n Huge thanks to our amazing team! 🙌. Grateful to my wonderful collaborators: @QianzhongChen, @XiaohanZhang220, @JiankaiSun.And to our incredible advisor @MacSchwager!.

💡5/n Real-world tests. We evaluate ParticleFormer on 3 real-robot tasks:.It consistently outperforms strong baselines in both dynamics prediction and downstream control!

💡4/n Hybrid Point Cloud Loss. Modeling dynamics from raw point clouds is tough — how do you capture both large motions and subtle deformations without point-level correspondences?. 🔍 We combine Chamfer Distance (local accuracy) with differentiable Hausdorff Distance (global

💡3/n Transformer-Based Dynamics Modeling. Previous particle-based world models (e.g., GNNs) rely on hand-crafted hyperparameters like max neighbors (TopK) and distance cutoffs. These hyperparameters are fragile — especially in contact-rich, multi-material scenes. 🧠

💡2/n Our key contributions:. ✅ A Transformer-based dynamics model that replaces GNN-style graphs with attention. ✅ A hybrid loss (Chamfer + Hausdorff) that supervises both local details & global structure — without expensive 3D reconstruction. ✅ An extended 3D dynamics

💡1/n Why does this matter?. 🔧 Generalizable manipulation requires models that decouple task logic from environment dynamics. 🧠 Humans do this naturally. But most robot models are still:. tuned to a single material or object.overly reliant on pre-defined topologies (e.g.,

🚀 Excited to share our #CoRL2025 paper! See you in Korea 🇰🇷!🎉. We present ParticleFormer, a Transformer-based 3D world model that learns from point cloud perception and captures complex dynamics across multiple objects and material types !. 🌐 Project website:

Unfortunately I cannot attend the conference in person this year, but our co-author @Kevin_GuoweiXu will be presenting the paper and answer all your questions!. 📜Poster session:.Time: Wed 16 Jul 11 a.m. PDT — 1:30 p.m. PDT.Location: West Exhibition Hall B2-B3 #W-607.

RT @sizhe_lester_li: Now in Nature! 🚀 Our method learns a controllable 3D model of any robot from vision, enabling single-camera closed-loo….

RT @agiachris: What makes data “good” for robot learning? We argue: it’s the data that drives closed-loop policy success!. Introducing CUPI….

RT @priyasun_: How can we move beyond static-arm lab setups and learn robot policies in our messy homes?.We introduce HoMeR, an imitation l….

RT @yjy0625: Introducing Mobi-π: Mobilizing Your Robot Learning Policy. Our method:.✈️ enables flexible mobile skill chaining.🪶 without re….

RT @tylerlum23: 🧑🤖 Introducing Human2Sim2Robot!  💪🦾. Learn robust dexterous manipulation policies from just one human RGB-D video. Our Rea….

Excited to share that MENTOR has been accepted to #ICML2025!.See you in Vancouver this July🤖.

RT @ZhengrongX: 𝑫𝒆𝒎𝒐𝑮𝒆𝒏 has been accepted to #RSS2025 🥳.See you in LA this June 🙌.

RT @GuanyaShi: When I was a Ph.D. student at @Caltech, @lschmidt3 discussed the paper "Do ImageNet Classifiers Generalize to ImageNet?" in….

RT @ju_yuanchen: 🍌We present DenseMatcher！.🤖️DenseMatcher enables robots to acquire generalizable skills across diverse object categories b….

RT @Kevin_GuoweiXu: 🚀 Introducing LLaVA-o1: The first visual language model capable of spontaneous, systematic reasoning, similar to GPT-o1….