Neurally Integrated Finite Elements for Differentiable Elasticity on Evolving Domains

Abstract

We present an elastic simulator for domains defined as evolving implicit functions, which is efficient, robust, and differentiable with respect to both shape and material. This simulator is motivated by applications in 3D reconstruction: it is increasingly effective to recover geometry from observed images as implicit functions, but physical applications require accurately simulating and optimizing-for the behavior of such shapes under deformation, which has remained challenging. Our key technical innovation is to train a small neural network to fit quadrature points for robust numerical integration on implicit grid cells. When coupled with a Mixed Finite Element formulation, this yields a smooth, fully differentiable simulation model connecting the evolution of the underlying implicit surface to its elastic response. We demonstrate the efficacy of our approach on forward simulation of implicits, direct simulation of 3D shapes during editing, and novel physics-based shape and topology optimizations in conjunction with differentiable rendering.

18 pages, 24 figures

References in corpus (12)

• Neural Dual Contouring
• Flexible Isosurface Extraction for Gradient-Based Mesh Optimization
• Topology Optimization Using Polytopes
• DiffTaichi: Differentiable Programming for Physical Simulation
• gradSim: Differentiable simulation for system identification and visuomotor control
• Differentiable solver for time-dependent deformation problems with contact
• Differentiable Simulation of Inertial Musculotendons
• PAC-NeRF: Physics Augmented Continuum Neural Radiance Fields for Geometry-Agnostic System Identification
• MeshFormer: High-Quality Mesh Generation with 3D-Guided Reconstruction Model
• PhyRecon: Physically Plausible Neural Scene Reconstruction
• GS-LRM: Large Reconstruction Model for 3D Gaussian Splatting
• Physically Compatible 3D Object Modeling from a Single Image