PhysAg3D Turns Photos Into Interactive 3D
PhysAg3D converts a single image into a textured 3D asset with articulated joints, deformable regions, and simulation-ready physical interactions. Its vision-language agent reasons about object parts and affordances, then iteratively refines the asset using geometric and physics tools.
PhysAg3D targets the biggest weakness in image-to-3D generation: visually convincing meshes that cannot meaningfully move, deform, or interact.
- –Separates visual generation from physical interaction construction, preserving the strengths of existing 3D generators
- –Supports both rigid articulation and compliant deformation, broadening use beyond robot joints
- –Uses rendered interactions and execution measurements for targeted, budgeted revisions
- –Reported evaluations show strong action completion and stability across real-world objects
- –Particularly promising for robotics, embodied-agent training, and simulation data generation
DISCOVERED
1h ago
2026-11-01
PUBLISHED
1h ago
2026-11-01
RELEVANCE
AUTHOR
AI Search
