Research
Structure from Collision via Hypothesis-Space Reasoning
This paper introduces Structure from Collision (SfC), a method to recover hidden interior cavities of objects from multiview collision observations. The authors reformulate SfC as hypothesis-space reasoning, using gradient-free optimization over explicit structural hypotheses with forward MPM simulation, followed by gradient-based refinement with differentiable MPM. The approach outperforms density-based baselines on the SfC benchmark and is evaluated on interior-topology stress tests and robustness analyses.
computer visionphysics simulationrobotic manipulation