HSImul3R: Physics-in-the-Loop Reconstruction of Simulation-Ready Human-Scene Interactions
This addresses the perception-simulation gap for embodied AI applications, allowing stable interactions in physics engines, though it is incremental in refining existing reconstruction methods with physical constraints.
The paper tackles the problem of reconstructing human-scene interactions from casual captures, where existing methods produce visually plausible but physically unstable results, and introduces HSImul3R, a framework that uses physics-in-the-loop optimization to achieve stable, simulation-ready reconstructions, enabling direct deployment to humanoid robots.
We present HSImul3R, a unified framework for simulation-ready 3D reconstruction of human-scene interactions (HSI) from casual captures, including sparse-view images and monocular videos. Existing methods suffer from a perception-simulation gap: visually plausible reconstructions often violate physical constraints, leading to instability in physics engines and failure in embodied AI applications. To bridge this gap, we introduce a physically-grounded bi-directional optimization pipeline that treats the physics simulator as an active supervisor to jointly refine human dynamics and scene geometry. In the forward direction, we employ Scene-targeted Reinforcement Learning to optimize human motion under dual supervision of motion fidelity and contact stability. In the reverse direction, we propose Direct Simulation Reward Optimization, which leverages simulation feedback on gravitational stability and interaction success to refine scene geometry. We further present HSIBench, a new benchmark with diverse objects and interaction scenarios. Extensive experiments demonstrate that HSImul3R produces the first stable, simulation-ready HSI reconstructions and can be directly deployed to real-world humanoid robots.