Digital Humans
Physics-based simulation of digital humans, spanning upsampled facial animations, interactive hair dynamics, and large-scale character clothing.
Overview
We are broadly interested in developing advanced algorithms and tools that embed more accurate physics into neural understandings and encodings of the world. Our research spans both traditional (forward) and inverse physics solvers, with applications in robotics, engineering analysis, and computer graphics. We prioritize high-fidelity results and close the gap to real-time performance through sparse 3D representations and neural acceleration.
Point of Contact:
Ken Museth
Key Research Areas
Finite Element & Monte Carlo Methods • Neural Solvers • Digital Humans • Synthetic Data Generation
Featured Research Projects
Finite Element Methods
GPU-accelerated finite element framework in Python and NVIDIA Warp, targeting diffusion, convection, fluid flow, and elasticity problems.
Neural Solvers
Mesh-free, geometry-agnostic, elastic simulation on any point-sampled geometry, such as point clouds, meshes, as well as 3D Gaussian splats.
Monte Carlo Simulation
GPU-accelerated, grid-free Monte Carlo solvers for static physics in complex 3D scenes, inspired by ray-tracing methods for photorealistic rendering.
Sparse Volumes
GPU-accelerated sparse volume data structures based on the OpenVDB toolset for large-scale, high-resolution 3D simulation and spatial intelligence.
Resources & Publications
Please visit our team's webpage for a full list of publications and resources.