Research questionHow can multimodal world models maintain physically coherent 3D scenes during closed-loop interaction?Interactive world models must keep geometry, appearance, physical behavior, and viewpoint changes mutually consistent over time. Modeling these elements separately can cause future views or reconstructions to drift. Latest papersRecent research connected to this question, newest first.TourPhysics: Bringing Physics to World Models for Exploration and Manipulation from a Single ImageThe source describes TourPhysics, an online framework initialized from a single image and declarative physical configuration. It combines deterministic simulation with video generation for simulator-defined camera tours and object manipulations; the reported evidence indicates closer adherence to prescribed trajectories, preservation of the input scene, and reduced appearance drift during long-horizon revisits compared with evaluated baselines.research paper · Sep 7, 2026Puffin-World: Scaling a Unified Multimodal Model with Native 3D World StatesThe paper presents a unified model that jointly represents physics, geometry, appearance, and camera state while propagating dynamics across future frames. It reports experiments involving 3D generation, reconstruction, future-view synthesis, and exploration, with evidence limited to the described system, data, and tasks.research paper · Sep 3, 2026