OfficeTwin: Realizing 3D Reconstruction of Redundant Scenes through Partial Viewpoint Perception

This research introduces OfficeTwin, a robot-native framework that converts incomplete observations from cameras mounted on robots with limited degrees of freedom into simulation-ready digital twins. By imposing metric and structural constraints and registering repeated office cubicles into a shared canonical frame, the system recovers occluded geometry without hallucination. On a real office scan, OfficeTwin successfully reconstructs every cubicle as an individually watertight, metric solid, outperforming feed-forward reconstruction and generative baselines that fail to produce simulation-ready assets under these conditions.