- Uri Ascher, Professor, University of British Columbia
- L. Mahadevan, Professor, Harvard University
- Mridul Aanjaneya, Assistant Professor, Rutgers University
- Chenfanfu Jiang, Assistant Professor, University of Pennsylvania
- Mélina Skouras, Research Scientist, Inria Grenoble Rhône-Alpes
- Steve Tonneau, Lecturer (Associate Professor), University of Edinburgh
- Etienne Vouga, Assistant Professor, UT Austin
- Sophie Jörg, Associate Professor, Clemson University
The many faces of stiffness
Uri Ascher, Professor, University of British Columbia
Abstract: The words “stiff”, “stiffness”, “stiffening”, etc., arise often in applications when simulating, calibrating and controlling dynamics. But these words often have different meanings in different contexts. A subset on which we will concentrate includes: (1)textbook-type (decaying) numerical ODE stiffness; (2) highly oscillatory stiffness; (3) stiffness matrix; and (4) numerical stiffening. Some of these terms are popular in scientific computing, while others come from mechanical engineering. A potential confusion may arise in this way, and it gets serious when more than one meaning is encountered in the context of one application. Such is the case with the simulation of deformable objects in visual computing, where all of the above appear in one way or another under one roof. In this lecture I will describe the meaning of stiffness in each of these topics, how they arise, how they are related, what practical challenges they bring up, and how these challenges are handled in context. The concepts and their evolution will be demonstrated. It is about meshes, their resolution and spectral properties.
BIO: Uri M. Ascher is a professor emeritus in the Department of Computer Science at the University of British Columbia. He obtained his PhD from the University of Minnesota in 1975. He is Fellow of the Royal Society of Canada, SIAM Fellow, and recipient of the CAIMS Research Prize, and has co-authored four textbooks and many scientific papers. The focus of his work is on the investigation, promotion and application of novel, efficient and reliable methods in scientific computation, and their application in computer graphics, imaging, inverse problems and other approximation problems involving differential equations with constraints and optimization.
Predicting and programming shape
L Mahadevan, Professor, Applied Mathematics, Physics, and Organismic and Evolutionary Biology, Harvard University
Abstract: Nearly a century since the publication of D’Arcy Thompson’s eponymous
classic, On growth and form, his vision has finally begun to permeate into the fabric of modern biology. Within this backdrop, I will show how a combination of biological and physical experiments, mathematical models and computations allow us to begin unraveling the physical basis for morphogenesis in the context of examples such as leaves, guts and brains. These questions also raise the possibility of designing shape for function in an engineering setting. I will describe our attempts to solve this problem in a few different settings that include origami and kirigami tessellations for complex surfaces, and 4d phytomimetic printing strategies, using a combination of experimental, computational and theoretical approaches.
The power of constitutive modeling in physics-based animation: elasticity, inelasticity, and damage mechanics
Chenfanfu Jiang, Assistant Professor, University of Pennsylvania
Abstract: Photorealistic and efficient simulation of natural phenomena in virtual worlds and characters is an essential aspect of computer graphics that remains challenging. The most challenging yet ubiquitous materials are those whose dynamics involve dramatic topological, stiffness, and phase changes, and those who undergo close contact, frequent collision, and sensitive coupling with their surroundings. Examples include flowing sand, breaking ice, tangled clothes, interweaving yarns, smashed fruits, and ripped-apart fibers. This talk aims to demonstrate that a lot of these superficially complicated material behaviors can be abstracted into simple constitutive modeling of elasticity and inelasticity. Using unified continuum solvers such as the Material Point Method, the physical modeling can be mostly or entirely separated from numerical discretization, allowing rapid and convenient exploration of simulating a new material and capturing a new effect for physics-based animation.
Bio: Chenfanfu Jiang is an Assistant Professor of Computer and Information Science at the University of Pennsylvania. He joined the department in 2017 after two-year’s postdoctoral research at UCLA, where he also received his Ph.D. degree under advisors Demetri Terzopoulos and Joseph Teran in 2015. He was awarded UCLA Edward K. Rice Outstanding Doctoral Student for his dissertation work. He is currently a faculty of Penn’s SIG Center for Computer Graphics, the Penn Institute for Computational Science, and the Graduate Group in Applied Mathematics and Computational Science. His research interests include physics-based simulation for computer graphics, computational solid and fluid mechanics, and high-performance scientific computing.
Bridging the gap between character animation and legged robotics through feasibility
Steve Tonneau, Lecturer (Associate Professor), University of Edinburgh
Abstract: The scientific issues of animating a virtual legged character or controlling a legged robot share the same long-term objective:
to provide a generic motion synthesis method, able to accurately achieve the largest variety of tasks, in an efficient and natural looking fashion. The input of such method would only require the definition of high-level goals, and thus would not require expert knowledge.
Acknowledging the proximity of both scientific domains, a slowly increasing number of researchers contribute to both communities.
This is not an easy task, as these fields significantly differ in their immediate objectives. In the absence of a universal motion synthesis method, different compromises for an acceptable method are considered.
For instance, unexpected collisions between a robot and its environment are currently unacceptable, while an animator will not be satisfied with a virtual character successfully achieving an unnatural looking grasping. Yet the opposite situations are usually tolerated. Unfortunately most motion synthesis methods do not offer flexibility in the choice of the criteria that define an acceptable solution. As a result, proposing contributions relevant for both communities is a challenging task.
Our research thus aims at providing holistic methods for motion synthesis, independent from their application. We propose to reformulate a standard motion synthesis problem into a feasibility problem. The scientific objective is changed from the computation of a motion that accomplishes a given task, into the characterization of all the motions that allow to accomplish that same task. The by-product of this characterization is then the set of all possible solutions, from which it is possible to select one optimizing the user-defined tradeoff.
In this talk we will formally define the locomotion feasibility problem for legged avatars and robots, and provide concrete implementations of this formalism that achieve a step towards a universal motion synthesis technique.
Bio: Steve Tonneau is a lecturer at the University of Edinburgh. He defended his Phd in 2015 after 3 years in the INRIA/IRISA Mimetic research team, and pursued a post-doc in robotics at LAAS-CNRS in Toulouse, within the Gepetto team. His research focuses on motion planning based on the biomechanical analysis of motion invariants. Applications include computer graphics animation as well as robotics.
Efficient Solver Design for Scalable Fluid Simulation with Structure Interaction
Mridul Aanjaneya, Assistant Professor, Rutgers University
Abstract: Iterative Krylov solvers are memory-efficient and produce approximate solutions even when terminated early, but can require many iterations with poorly conditioned matrices, or strongly-coupled systems. In contrast, direct solvers avoid these issues, but can incur a significant memory overhead. By utilizing a carefully-crafted hybrid approach rooted in the principles of elliptic partial differential equations, this talk will show that it is possible to obtain the best of both worlds while avoiding most of their drawbacks, outperforming existing state of the art solutions. Our focus will be on strongly-coupled (or monolithic) formulations for multi-material flows, possibly with solid-fluid coupling. We will show that our hybrid approach can be used to extend the applicability of fast Multigrid solvers for incompressible fluids to a much broader class of problems, which commonly arise in many computer graphics applications.
Bio: Mridul Aanjaneya is an Assistant Professor in the Department of Computer Science at Rutgers University since Fall 2017, where he is a faculty member in the Computational Biomedicine Imaging and Modeling Center (CBIM). Prior to joining Rutgers, he was a postdoctoral researcher at the University of Wisconsin – Madison, where he was advised by Eftychios Sifakis. He received his PhD in Computer Science from Stanford University in June 2013 under the supervision of Ron Fedkiw. His research interests are in computer graphics, physics-based simulation, scientific and high-performance computing, and their applications in the physical sciences and engineering.
On the design of physics-aware inverse modeling tools for digital fabrication
Mélina Skouras, Research Scientist, Inria Grenoble Rhône-Alpes
Abstract: Recent technological advances in digital manufacturing enable experts and casual users alike to design and fabricate complex structures with a fine control over their external geometry and internal material distribution. However, designing an object that is truly functional remains a challenging task. This is because, in the real world, the object is subject to physical laws that affect its behavior, e.g. determine its final shape. Therefore, the designer needs to estimate and invert the effects of the physics on the object to obtain a structure that behaves as desired. In this talk, I will explain how inverse modeling tools can be used to automate the design of custom elastic objects with prescribed target shapes. I will focus on some important challenges that arise when designing such tools and present the principles that we followed to address them when working on concrete examples.
Bio: Mélina Skouras is a tenured researcher at Inria Grenoble – Rhône-Alpes, France. Prior to joining Inria in December 2017, she was a postdoctoral associate at MIT where she worked with Wojciech Matusik on the computational design of meta-materials. She obtained her PhD in 2014 from the Computer Graphics Laboratory of ETH Zurich, Switzerland, in collaboration with Disney Zurich, under the supervision of Markus Gross. Her thesis focused on the development of novel algorithms for the design of custom deformable objects. She received her Master’s degree in Computer Science and Applied Mathematics from ENSIMAG, INP Grenoble, France, in 2004. Before starting her PhD, she was a software developer at Dassault Systèmes, in the CATIA Geometric Modeler team.
How to synthesize subtle eye motions based on perceptual findings
Sophie Jörg, Associate Professor, Clemson University
Abstract: Eye motions constitute an important component of animating compelling virtual characters and agents. They can convey a person’s thoughts, emotions, and mood, and they play a vital role in face-to-face interactions. From an animation point of view, eye motions consist of several elements: the eyeball motions, which entail fixations and saccades; the changes in size of the pupil diameter responding, for example, to light intensity changes; and periocular motions mainly consisting of blinks and lid saccades. However, our eyes perform further motions such as microsaccades, ocular drift, ocular microtremor, and pupil unrest, which are typically not considered in animation.
In this talk, I will discuss how we can synthesize more compelling eye animations, with a focus on subtle eye motions, based on results from a series of perceptual experiments. Our results show that adding a small amount of synthesized jitter to the eyeball rotation and pupil diameter of virtual characters increases their perceived naturalness.
Bio: Sophie Jörg is an Associate Professor in the School of Computing at Clemson University. Her research is in the areas of character animation, motion perception, virtual reality, and human computer interaction, with a special interest in hand motions. She received an NSF CAREER award in 2017. Previously, she worked as a visiting researcher and postdoctoral researcher at Carnegie Mellon University, as an intern at Disney Research, Pittsburgh, and as a junior researcher at Fraunhofer IAIS. She holds a PhD from Trinity College Dublin.
Simulating Multi-body Collisions
Etienne Vouga, Assistant Professor, University of Texas at Austin
Abstract: Correctly resolving contact and impact between colliding bodies remains one of the most challenging simulation tasks. Not only is robust collision response important for computer animation, but as the computer graphics community continues to tackle inverse problems in 3D printing, mechanism design, and computational architecture, there is a rising need to formulate concise and correct contact constraints. Despite decades of research, a wide gulf remains between practical algorithms for handling collisions, which typically layer impulses and penalty forces and failsafes, and the ideal contact response algorithm that provably maintain non-interpenetration of the simulated objects, without falling prey to “Zeno-type” paradoxes, infinite loops, or artificial sticking or damping. I will give an overview of why computational contact mechanics is so hard, especially for multiple colliding bodies; describe some recent work analyzing the safety of contact response algorithms, and present some counterexamples, open problems, and potential future directions related to modeling frictional contact.
Bio: Dr. Etienne Vouga is an assistant professor of computer science at the University of Texas at Austin. He received his Ph.D. at Columbia University in 2013 under Eitan Grinpsun and spent a year as an NSF Mathematical Sciences Postdoctoral Fellow at Harvard, working with L. Mahadevan. His research is at the intersection of computer graphics, computational mechanics, and mathematics, and his specific interests include physical simulation of everyday materials like cloth, hair, and paper; algorithms for detecting and resolving collisions that arise during animations; and using geometry processing algorithms and discrete differential geometry to solve design problems such as creating masonry buildings that stand up under their own weight, or origami patterns that compact into a small volume and deploy into large curved shapes.