Fast But Accurate: A Real-Time Hyperelastic Simulator with Robust Frictional Contact

Ziqiu Zeng, Siyuan Luo, Fan Shi, Zhongkai Zhang · 2025

Read Fast But Accurate: A Real-Time Hyperelastic Simulator with Robust Frictional Contact by Ziqiu Zeng, Siyuan Luo, Fan Shi, Zhongkai Zhang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-15078.txt.

Introduction

Simulating soft, stretchy materials in real time, with contact and friction, is one of the hard problems of computer graphics. Zeng and co-authors present a simulator that handles hyperelastic objects at interactive speeds without giving up accuracy.

Their framework runs on the GPU and folds the rules of non-penetration and friction into a local-global solver as complementarity conditions, which brings rapid convergence. A sparse way of handling the system inverse keeps the math parallel-friendly, and a splitting trick for the non-smooth parts sharpens the friction model. Tests ranged from large scenes and extreme stretching to precise friction, across soft and stiff materials alike. The authors stress that the method stays simple, resting on standard matrix operations.

Catalog record

CREATORZiqiu Zeng, Siyuan Luo, Fan Shi, Zhongkai Zhang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE96 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-15078

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Fast But Accurate: A Real-Time Hyperelastic Simulator with Robust Frictional Contact Ziqiu Zeng, Siyuan Luo, Fan Shi, Zhongkai Zhang Abstract . We present a GPU-friendly framework for real-time implicit simulation of elastic material in the presence of frictional contacts. The integration of hyperelasticity, non-interpenetration contact, and friction in real-time simulations presents formidable nonlinear and non-smooth problems, which are highly challenging to solve. By incorporating nonlinear complementarity conditions within the local-global framework, we achieve rapid convergence in addressing these challenges. While the structure of local-global methods is not fully GPU-friendly, our proposal of a simple yet efficient solver with sparse presentation of the system inverse enables highly parallel computing while maintaining a fast convergence rate. Moreover, our novel splitting strategy for non-smooth indicators not only amplifies overall performance but also refines the complementarity preconditioner, enhancing the accuracy of frictional behavior modeling. Through extensive experimentation, the robustness of our framework in managing real-time contact scenarios, ranging from large-scale systems and extreme deformations to non-smooth contacts and precise friction interactions, has been validated. Compatible with a wide range of hyperelastic models, our approach maintains efficiency across both low and high stiffness materials. Despite its remarkable efficiency, robustness, and generality, our method is elegantly simple, with its core contributions grounded solely on standard matrix operations. Figure 1 . Crossing Gingerbread Man: we propose a new framework for stable simulation of hyperelastic materials with nodes under large deformation and generic contact constraints in real-time. Pulling the gingerbread man ( 58.5k DoFs for a single object) through the thin and irregular obstacles is simulated at 11.95ms /iteration using 5 local-global iterations per frame when maximum contact pairs are involved ( 800 contact constraints). This is the teaser figure for the article. 1. Introduction Physics-based simulation of deformable materials plays a crucial role across a variety of disciplines, such as computer graphics, robotics, and medical imaging. Within these fields, it enhances visual experiences through accurate simulations of soft objects and their interactions with the environment, often utilizing finite element discretization for complex nonlinear materials. An ideal physical simulator should possess key attributes like the capacity for generality in simulating different materials and the robustness to minimize failure instances. In interactive applications, achieving real-time performance is essential, while a framework’s simplicity is highly valued for simplifying maintenance and development processes, making it widely applicable across various downstream fields. Position-based dynamics (PBD) ( Müller et al., 2006 ) has gained widespread adoption in real-time simulation engines due to its simplicity, robustness, and efficiency. However, as PBD is not derived from continuum mechanics principles, its real-time performance comes at the cost of accuracy. To address this limitation, Projective Dynamics (PD) ( Bouaziz et al., 2014 ) employs local-global optimization, combining localized constraint solving with global optimization for better accuracy and efficiency. Vertex Block Descent (VBD) ( Chen et al., 2024 ) further enhances real-time performance by optimizing matrix structures and constraints for parallelization, delivering significant computational advantages. In multi-body systems, addressing contact problems typically involves utilizing either penalty methods with soft constraints or Lagrange multiplier methods with hard constraints. While penalty methods are computationally efficient, they encounter difficulties in enforcing strict non-penetration and accurate friction. Incremental Potential Contact (IPC) ( Li et al., 2020 ) mitigates this using a logarithmic barrier penalty and barrier-aware line search to prevent intersections. Lagrange multiplier methods are robust and accurate but face challenges with convergence and parallelization in relaxation techniques like projected Gauss-Seidel ( Duriez, 2013 ) . Non-smooth Newton methods ( Macklin et al., 2019 ) , combined with Krylov subspace solvers and complementarity preconditioner, offer rapid convergence, high accuracy, and efficient parallelization. We propose a framework where the local-global iterations are constrained by nonlinear complementarity conditions. Our simulator, empowered by efficient matrix operations, exhibits many favorable features: First, our framework is efficient and enables large-scale simulations in real-time, in the presence of nonlinear and non-smooth problems. Within the real-time computational constraints, our solver is able to converge to a desirable accuracy , thereby enhancing the stability . Second, our method is not only general for a wide range of hyperelastic models but also preserves efficiency across materials with both low and high stiffness. Through extensive experiments, our framework has been proven to be robust when solving the complex contact problem with different challenges, including large deformation, non-smooth contact, and accurate friction. Beyond these advantages, our method shines in its simplicity, as the core contributions only rely on standard matrix operations. In summary, our contributions are listed as follows: (1) Our method achieves a highly parallelizable structure while maintaining a high convergence rate by transforming the linear global system into sparse matrix multiplications. (2) We reformulate Lagrange multiplier methods, particularly the non-smooth Newton method, to integrate seamlessly with local-global integrators. (3) We introduce a strategy to separate non-smooth indicators, resulting in both reduced Schur-complement computations and enhanced complementarity preconditioner. (4) Our algorithm establishes a unified, GPU-friendly system for real-time dynamics and contact resolution, with a modular design that ensures easy integration with other frameworks. 2. Related work (a) Frame 0 (b) Frame 1100 (c) Frame 3200 Figure 2 . Grabbing Raptor: stable catching of an elastic raptor with a soft gripper actuated by cables. Lifting, rotating, and moving the raptor by the fingers are complex operations where friction constraints are necessary. Physics-based simulation in computer graphics has been extensively studied. In this section, we focus on discussing the most recent work related to our method. 2.1. Implicit simulation for elastic dynamics In computer graphics, implicit simulations ( Bro-Nielsen and Cotin, 1996 ) using backward Euler integration allow for larger time steps compared to explicit methods ( Comas et al., 2008 ) , significantly improving computational stability in stiff systems. Finite element (FE) models ( Sifakis and Barbic, 2012 ; Kim and Eberle, 2022 ) are a valuable tool for understanding the underlying mechanisms in the real world, as they provide a direct explanation of soft tissue behavior through constitutive relations. With the rapid advancement of computational power and methods, FE models have become increasingly suitable for real-time and interactive simulations. In terms of integration, the traditional implicit method employs the Newton iteration technique to solve nonlinear problems. Initially restricted to linear elastic models ( Bro-Nielsen and Cotin, 1996 ) , the method is later extended to the co-rotational formulation ( Felippa, 2000 ) and hyperelastic and viscoelastic materials ( Marchesseau et al., 2010 ) . Although the Newton’s method converges fast in solving nonlinear systems, it involves re-evaluate and invert the Hessian matrix in each Newton iteration, which implies large computing costs. In practice, Newton integrators typically perform only one iteration in real-time applications ( Faure et al., 2012 ) . In computer graphics, PBD ( Müller et al., 2006 ) is very popular in applications like cloth simulations owning to its high efficiency and stable behavior. As an extension, the extended PBD (XBPD) ( Macklin et al., 2016 ) addresses the limitation of stiffness dependence, providing a better approximation to the implicit Euler method. Recently, the VBD method ( Chen et al., 2024 ) is proposed, which constrains the positions with hyperelastic constraints. We categorize these methods as PBD-like methods ( Bender et al., 2017 ) , where the positions of the

Read the complete text