Recovering Projective Dynamics: Strain‐Based ADMM for Cosserat Rod Simulation
Abstract The Cosserat rod model captures the rich dynamics of slender structures yet is costly to simulate. Projective Dynamics (PD) accelerates simulation via a prefactorized system matrix but converges poorly under the extreme stretch‐shear to bend‐twist stiffness ratios of realistic materials, forcing practitioners to soften stiffness at the cost of visual artifacts. Alternating Direction Method of Multipliers (ADMM)‐based extensions handle this nonlinearity at the cost of local optimization overhead, yield limited acceleration, and exhibit perceptible damping under tight iteration budgets. In this paper, we propose an efficient and robust simulation framework based on a strain‐based ADMM formulation. By defining auxiliary variables at the strain level and employing a two‐level mathematical relaxation via Block Coordinate Descent and Projective Dynamics, our method recovers the fast, prefactorizable local‐global architecture of classic PD while avoiding nonlinear local solves. We further augment the orientation block of the system matrix for bending‐twisting constraints and introduce a robust penalty parameter clamping strategy. As a result, our method robustly simulates trees and hair under physically realistic parameters without per‐scene tuning of the elastic penalty. Experiments show that our method converges faster than previous ADMM variants, reduces total runtime by 21% in a representative high stiffness ratio scenario, and exhibits less numerical damping than prior ADMM variants under tight iteration budgets.
Authors
- Kangrui Zhang (ORCID: https://orcid.org/0009-0004-8754-3995)
- Bo Ren (ORCID: https://orcid.org/0000-0001-8179-9122)
Institutions
- Nankai University (CN)
Publication Details
- Journal
- Computer Graphics Forum
- Published
- 2026-11-13
- DOI
- https://doi.org/10.1111/cgf.70641
- Primary Topic
- Computer Graphics and Visualization Techniques
- Type
- article
- Field-Weighted Citation Impact
- 0.00