CRII: CHS: Capturing Emergent Fine-Scale Features in Visual Simulation of Elasticity

自然现象的视觉模拟是数字动画中的一个既定工具，在设计和培训应用中也越来越重要。特别是对织物和软组织等弹性可变形对象的交互模拟，对于服装设计、电子商务和医疗培训具有直接好处，在这些领域，它可以高效和准确地预览服装设计或手术操作。但要让模拟技术在这些应用中真正有价值(以便它们影响现实生活中的决策)，它们必须在视觉上代表真实现象的行为，包括出现的细微特征，如皱纹和褶皱，这些特征必须在传达有关材料潜在物理属性的信息时可靠而忠实地再现。由于直接的高分辨率仿真对于可视化应用来说代价太高，因此已经开发了许多技术来高效地表示细节，包括自适应细化和过程合成。这两种方法都取决于能否确定当前的模拟分辨率是否不足以解决系统中应该存在的细微尺度细节。要做到这一点，人们必须预测这样的细节可能在何时出现，以及它将如何随着时间的推移而演变。遗憾的是，现有的标准要么依赖于昂贵的后验误差估计，要么依赖于需要仔细手动调整参数以产生可靠结果的启发式方法。这项研究的目的是朝着健壮和通用的基于物理的技术的目标取得进展，该技术可以准确地预测未分辨的细尺度特征的形成并在模拟进行时表征它们的行为。这项工作中开发的用于细节预测的通用框架将使视觉模拟更有效、更可靠，并更能代表所描述现象的真实行为，从而促进其在服装设计、手术规划和培训等应用中的采用。PI将通过与他所在大学的设计系和外科医生合作，将项目成果转移到这两个领域的从业者；这将通过与设计和外科专家的互动，额外培训计算机科学专业的学生进行跨学科研究。这项工作在视觉艺术、设计和医学中的应用将被用于拓展，并吸引原本不会被STEM领域吸引的艺术和设计专业的学生。这个项目将解决数值模拟中的一个基本挑战：如何有效地描述模拟物理系统中未解决的细节？PI以前的工作指出，动力不稳定性是预测细尺度特征出现的基本标准。这项工作将建立一个理论和计算框架，将这一见解形式化，并提供有效和实用的技术来捕捉可变形物体模拟中的紧急细节。这项工作的贡献将是一个新的数值方法的框架，可以在一般情况下自动检测和表征动力不稳定性。这样做将扩展先验细化技术的多功能性，使其适用于广泛类别的弹性模拟问题。所提出的方法将在服装设计和虚拟手术等应用中出现的实际问题上进行评估。所获得的知识将促进对动态系统仿真中定量和定性保真度的理解，并将为今后的自适应仿真工作提供良好的理论基础。

Visual simulation of natural phenomena is an established tool in digital animation and is also growing in importance in design and training applications. Interactive simulation of elastic deformable objects such as fabric and soft tissue, in particular, has a direct benefit for apparel design, e-commerce, and medical training, where it can enable efficient and accurate previewing of garment designs or surgical actions. But for simulation techniques to be truly valuable in these applications (so that they influence real life decisions), they must be visually representative of the behavior of the real phenomenon, including emergent fine-scale features such as wrinkles and folds, which must be reliably and faithfully reproduced as they convey information about the underlying physical properties of the material. Because direct high-resolution simulation is too expensive for visual applications, many techniques have been developed to represent the details efficiently, including adaptive refinement and procedural synthesis. Both types of methods depend crucially on being able to determine whether the current simulation resolution is inadequate to resolve the fine-scale detail that should exist in the system. To do so, one must predict when such detail is likely to emerge, and how it will evolve over time. Unfortunately, existing criteria for doing this rely either on expensive a posteriori error estimation, or on heuristic approaches requiring careful manual parameter tuning to generate reliable results. This research aims to make progress towards the goal of robust and versatile physics-based techniques that can accurately predict the formation of unresolved fine-scale features and characterize their behavior as the simulation proceeds. The versatile framework for detail prediction developed in this work will make visual simulation more efficient, reliable, and representative of the true behavior of the depicted phenomenon, stimulating its adoption in applications such as apparel design and surgical planning and training. The PI will transfer project outcomes to practitioners in both these fields through collaboration with design faculty and surgeons in his University; this will additionally train computer science students in interdisciplinary research through interactions with experts in design and surgery. Applications of this work in visual arts, design, and medicine will be used for outreach, and to engage art and design students who would not otherwise be attracted to STEM fields.This project will address a fundamental challenge in numerical simulation: how can one efficiently characterize unresolved detail in a simulated physical system? Previous work by the PI has pointed to dynamical instabilities as an essential criterion to predict the emergence of fine-scale features. This work will build a theoretical and computational framework that formalizes this insight, and provides efficient and practical techniques for capturing emergent detail in the simulation of deformable objects. The contribution of the work will be a framework of novel numerical methods that can automatically detect and characterize dynamical instabilities in a general setting. Doing so will extend the versatility of a priori refinement techniques, making them applicable to a broad class of elastic simulation problems. The proposed methods will be evaluated on practical problems arising in applications such as garment design and virtual surgery. The knowledge gained from this work will advance understanding of quantitative and qualitative fidelity in simulations of dynamical systems, and will provide a sound theoretical foundation for future work on adaptive simulation.