Brief

Researchers from Nanyang Technological University (NTU) have developed PhysX-Anything, a system capable of generating physically simulated 3D assets from a single image. This advancement moves beyond merely creating visually realistic 3D models to producing assets with accurate structural components, joint relationships, material properties, and functional behaviors. The system aims to significantly reduce the cost and effort associated with creating 3D assets for robotics training, AR/VR applications, and industrial simulations by automating the inference of physical attributes from a single input image. AI

Researchers have developed a new few-shot method called ATOP (Articulate That Object Part) that uses text prompts and motion personalization to animate static 3D objects. This approach leverages diffusion models to generate plausible motion samples, which are then personalized to a specific input 3D object. The method fine-tunes diffusion models to learn unique motion identifiers and then transfers this personalized motion to the 3D space for articulation parameter optimization. Experiments show ATOP achieves higher accuracy and better generalization in few-shot scenarios compared to existing methods. AI