🎥 Video Generation Enables Zero-Shot Robotic Manipulation 🤖 Introducing PhysWorld, a framework that bridges video generation and robot learning through (generated) real-to-sim world modeling. 🌐 Project: https://t.co/9mRqPqr5TS 📄 Paper: https://t.co/wmkEpmUGhq 💻 Code:

Good way to use video-generation world model in robotics

Generative video is becoming a new form of simulation. PhysWorld links video generation with robot learning, turning visual synthesis into real-to-sim modeling where zero-shot manipulation starts to emerge.

How can we turn a generated video into a robotic demonstration? Check out @PointsCoder 's recent work PhysWorld. We also open-sourced the whole pipeline that hopefully can make real-to-sim simpler.

Great work from our student researcher Jiageng Mao @PointsCoder to enable scalable robot learning by imitating AI-generated videos.

This is the most impressive world model -> physical AI training project I have seen published. I know world models are going to be a large part of solving the simulation data gap, this this really puts all of the pieces together. #Robotics #Simulation

This work is led by Jiageng as a student researcher project @GoogleDeepMind , collaborated with @SichengHe12345 , Hao-Ning Wu, Yang You, @Kevin_SSY, Zhicheng Wang, Yanan Bao, Huizhong Chen, @GuibasLeonidas, @vitorguizilini, and @howardzzh, and is advised by @yuewang314.

🔍 What did we find? By coupling video generation with physical world modeling, PhysWorld transforms purely visual signals into physically feasible actions. ✅ Enables zero-shot real-world manipulation ✅ Improves success rate by +15% over prior video-imitation methods ✅

💡 What is PhysWorld? PhysWorld enables robots to learn manipulation skills without real-world demonstrations. Given just one image and a task prompt, it: 1️⃣ Generates a task-conditioned video showing how to complete the task 2️⃣ Reconstructs a physically interactable 3D scene

Unified multimodal models can generate text and images, but can they truly reason across modalities? 🎨 Introducing ROVER, the first benchmark that evaluates reciprocal cross-modal reasoning in unified models, the next frontier of omnimodal intelligence. 🌐 Project:

Emily is presenting her first-ever paper at #ICCV2025. Welcome to come and have a chat with her!

🚀 Introducing InstantSfM: Fully Sparse and Parallel Structure-from-Motion. ✅ Python + GPU-optimized implementation, no C++ anymore! ✅ 40× faster than COLMAP with 5K images on single GPU! ✅ Scales beyond 100 images (more than VGGT/VGGSfM can consume)! ✅ Support metric scale.

Check out our new humanoid whole-body manipulation dataset!

Check out our work on leveraging Internet images for robotic msnipulation!

🚀 Join Us: Research Internships in Embodied Intelligence The USC Geometry, Vision, and Learning Lab ( https://t.co/MP3PFbYx2L) is seeking highly motivated interns to push the frontiers of AI, robotics, and 3D computer vision. You’ll work on large-scale VLA models,

This project is co-led by our incredible intern Wei Chow and me, and I am especially grateful to my advisor, @yuewang314 , for his invaluable guidance and support throughout this work. 🙏 We also deeply appreciate the contributions and insights of @Boyiliee, @DanielSeita, and

How do we fix this? Introducing PhysAgent 🚀 – a new framework that enhances VLMs by integrating: 🔹 Vision foundation models (Depth, SAM, GroundingDINO) 🔹 A physics knowledge memory for improved reasoning 🔹 Chain-of-thought inference for self-verification PhysAgent boosts

What did we find? 🧐 We evaluated 75 top VLMs, including GPT-4o, Gemini, and open-source models, and found: ✅ Strong commonsense reasoning but poor physical reasoning ✅ Closed-source models outperform open-source ones, but still struggle ✅ Scaling data and model size does not

What is PhysBench? PhysBench is a comprehensive benchmark with 10,002 video-image-text entries that assess VLMs across four major domains: 1️⃣ Physical object properties (number, mass, stiffness, elasticity, etc.) 2️⃣ Physical object relationships (distances, depths, velocities,

Can Vision-Language Models (VLMs) truly understand the physical world? 🌍🔬 Introducing PhysBench – the first benchmark to evaluate VLMs’ understanding of physics! PhysBench is accepted to #ICLR2025 as an Oral presentation (only 1.8% out of 11k submissions)! 🌐 Project: