자료유형  

 학위논문

Control Number  

0017164285

International Standard Book Number  

9798342138611

Dewey Decimal Classification Number  

646

Main Entry-Personal Name  

Jin, Yongxu.

Publication, Distribution, etc. (Imprint  

[S.l.] : Stanford University., 2024

Publication, Distribution, etc. (Imprint  

Ann Arbor : ProQuest Dissertations & Theses, 2024

Physical Description  

85 p.

General Note  

Source: Dissertations Abstracts International, Volume: 86-04, Section: B.

General Note  

Advisor: Fedkiw, Ron.

Dissertation Note  

Thesis (Ph.D.)--Stanford University, 2024.

Summary, Etc.  

요약Physics-based simulation techniques have long been established for various offline applications, yet real-time dynamic simulation remains a formidable challenge. Despite advancements in modern game engines like Unreal Engine 5, achieving high-resolution, real-time simulation capabilities remains limited. Recently, researchers have shown interest in using neural networks to approximate dynamic simulation, thanks to their fast inference on GPUs. However, although effective at approximating kinematics or quasistatic simulation, purely data-driven approaches often struggle with dynamic simulations (involving velocity and momentum information) due to potential overfitting and poor generalization with time series data. This limitation makes them unsuitable for real-world applications. Other efforts have tried using neural networks to upsample real-time, low-resolution simulations, but achieving good low-resolution results with conventional methods is very challenging.This thesis aims to pioneer a paradigm for real-time, high-fidelity physics simulation, with a focus on practical implementation within current game engines. Motivated by recent advancements in neural networks for capturing quasistatic simulations (referred to as quasistatic neural networks, or QNNs), we propose to rethink the need for dynamic components given QNN-based enhancements and redesign real-time physics models to primarily capture the ballistic motion of full dynamics, which can then be enhanced by QNNs to obtain the full shape. These meticulously designed physics models ensure stability, robustness, and performance that surpass real-time requirements. Concurrently, the lightweight QNNs can capture quasistatic shapes, facilitating ease of training and robust generalization. This thesis comprises two primary papers: one addressing human flesh simulation and the other focusing on the simulation of loose-fitting clothing.

Subject Added Entry-Topical Term  

Clothing.

Subject Added Entry-Topical Term  

Computer & video games.

Subject Added Entry-Topical Term  

Physics.

Subject Added Entry-Topical Term  

Success.

Subject Added Entry-Topical Term  

Ordinary differential equations.

Subject Added Entry-Topical Term  

Neural networks.

Subject Added Entry-Topical Term  

Bones.

Subject Added Entry-Topical Term  

Computer science.

Subject Added Entry-Topical Term  

Mathematics.

Added Entry-Corporate Name  

Stanford University.

Host Item Entry  

Dissertations Abstracts International. 86-04B.

Electronic Location and Access  

로그인을 한후 보실 수 있는 자료입니다.

Control Number  

joongbu:657671