Original title: Learning Flexible Body Collision Dynamics with Hierarchical Contact Mesh Transformer
Authors: Youn-Yeol Yu, Jeongwhan Choi, Woojin Cho, Kookjin Lee, Nayong Kim, Kiseok Chang, ChangSeung Woo, Ilho Kim, SeokWoo Lee, Joon Young Yang, Sooyoung Yoon, Noseong Park
This article discusses the development of a new graph neural network model called the Hierarchical Contact Mesh Transformer (HCMT) for modeling complex physical systems. While many existing models have been successful in reducing solving time and improving solution accuracy in fluid and rigid body dynamics, there has been limited exploration of their effectiveness in addressing the challenges of flexible body dynamics, specifically instantaneous collisions. HCMT aims to fill this research gap by utilizing hierarchical mesh structures that allow for learning long-range dependencies between spatially distant positions of a body. It consists of two components: the contact mesh Transformer (CMT) and the hierarchical mesh Transformer (HMT), which enable long-range interactions and quick propagation of collision effects to distant positions. Additionally, a new dataset reflecting experimental settings commonly used in product design is introduced, and the performance of HCMT is compared to other baseline models using benchmark datasets. The results demonstrate that HCMT outperforms existing methods in terms of performance improvements.
Original article: https://arxiv.org/abs/2312.12467