Collision Resolution for Cloth Simulations Using Machine Learning
Abstract
A key aspect of 3D animation is physics based simulation. To create realistic animations for objects such as cloth and fabric, a category of mathematical models known as softbody simulations are used. There are a wide variety of techniques for simulating the behavior of cloth, with varying degrees of performance and visual fidelity. One such technique that this project builds off of involves calculating the frictional forces between a cloth and intersecting geometries. This ensures that cloth objects display realistic behavior when sliding over rigid surfaces or being dragged by high friction materials. It is especially useful for depicting the movement of garments, as this technique can capture how real world clothing clings to the human form. With this technique, as with many, there is a tradeoff between performance and visual realism. While the latter is desirable for many applications in filmmaking and video game development, it can come at the cost of rendering time or real time performance. This friction based simulation produces visually realistic results, but it can be computationally expensive to simulate, especially for complex geometries. This project aims to create a deep learning model that can approximate the results of friction based collision resolution with better performance. By training a model on data collected from the real results of this technique, it could give similar results at potentially greater performance.
Citation
Yahya, Ahsan (2023). Collision Resolution for Cloth Simulations Using Machine Learning. Undergraduate Research Scholars Program. Available electronically from https : / /hdl .handle .net /1969 .1 /200251.