Abstract
Creating virtual digital replicas from real-world data unlocks significantpotential across domains like gaming and robotics. In this paper, we presentDRAWER, a novel framework that converts a video of a static indoor scene into aphotorealistic and interactive digital environment. Our approach centers on twomain contributions: (i) a reconstruction module based on a dual scenerepresentation that reconstructs the scene with fine-grained geometric details,and (ii) an articulation module that identifies articulation types and hingepositions, reconstructs simulatable shapes and appearances and integrates theminto the scene. The resulting virtual environment is photorealistic,interactive, and runs in real time, with compatibility for game engines androbotic simulation platforms. We demonstrate the potential of DRAWER by usingit to automatically create an interactive game in Unreal Engine and to enablereal-to-sim-to-real transfer for robotics applications.