UCLA

This thesis explores research advancements in building accurate physics-integrated digital twins. First, as the core computational engine, accurate and efficient simulations are essential for achieving realistic dynamics. To this end, we develop improved inelasticity simulation frameworks for both the Material Point Method (MPM) and the Finite Element Method (FEM). Additionally, we introduce FEM-MPM coupled simulation frameworks that leverage the strengths of both methods to enable multi-material simulations. Furthermore, we propose efficiency enhancements for cloth and inelasticity simulations. Second, digital twins must be grounded in real-world data. Leveraging recent advancements in neural rendering and differentiable rendering, we propose methods for estimating physical parameters from multiview videos, directly simulating reconstructed environments, immersively interacting with reconstructed worlds, and faithfully reconstructing simulation-ready garments from single-view images. Finally, we demonstrate the critical role of digital twins in mechanical design. Using our simulation-driven approach, we successfully fabricate an aerial vehicle with dual operational modes.