Advanced GPU techniques for dynamic remeshing and self-collision handling in real-time cloth tearing
Jong-Hyun Kim* Jung Lee**
(* : Inha University, ** : Hanbat National University)
IEEE Access 2025
Jong-Hyun Kim* Jung Lee**
(* : Inha University, ** : Hanbat National University)
IEEE Access 2025
Abstract : In this paper, we propose a GPU-based framework to accelerate dynamic remeshing and self-collision handling required for tearing thin-shell materials such as cloth: 1) Dynamic remeshing involves updating the shape and connectivity information at the element/node level, making parallel computing challenging and computationally costly. In this paper, we propose a new data structure that enables GPU-friendly calculations. Additionally, 2) to efficiently handle the computationally intensive self-collision handling in tearing cloth, we propose a GPU-based BVH (Bounding Volume Hierarchy) tree construction, update, and traversal method, along with an optimized R-Triangle (Representative-Triangle) technique. We also present a method to optimize kernels based on a complete binary tree in arbitrary triangular meshes, improving performance. Thin-shell materials are utilized in various fields including simulation/animation based on physics, games, surgical simulation, and virtual reality. Tearing cloth requires dynamically updating its geometry and connectivity, which is complex and computationally intensive. This process needs to be performed quickly, especially when dealing with interactive content. Most existing approaches perform remeshing through low-resolution simulations to maintain real-time performance or use pre-divided patterns, which cannot be considered dynamic remeshing and result in low-quality torn surface patterns. In this study, we propose a GPU-optimized dynamic remeshing algorithm that enables real-time processing of high-resolution cloth tearing. Additionally, we introduce a method to accelerate the computationally intensive self-collision handling in tearing cloth simulations using GPUs. To achieve this, we propose a GPU-based complete binary tree method to efficiently construct, update, and traverse a BVH (Bounding Volume Hierarchy) tree and optimize the R-Triangle (Representative-Triangle) technique on triangular meshes using GPUs to minimize primitive collision checks. The proposed method performs dynamic remeshing rather than using predefined split mesh structures, making it suitable for interactive physics-based modeling in applications such as virtual reality surgery simulation, real-time gaming, and virtual environments.
[paper]