SCA 17: Eurographics/SIGGRAPH Symposium on Computer Animation

Permanent URI for this collection

https://diglib.eg.org/handle/10.2312/2631998

Browse

Now showing 1 - 17 of 17

Rigid Body Contact Problems using Proximal Operators
(ACM, 2017) Erleben, Kenny; Bernhard Thomaszewski and KangKang Yin and Rahul Narain
Iterative methods are popular for solving contact force problems in rigid body dynamics. They are loved for their robustness and surrounded by mystery as to whether they converge or not. We provide a mathematical foundation for iterative (PROX) schemes based on proximal operators. This is a class of iterative Jacobi and blocked Gauss-Seidel variants that theoretically proven always converge and provides a flexible plug and play framework for exploring di erent friction laws.We provide a portfolio of experience for choosing r -Factor strategies for such schemes and we analyze the distribution of convergence behaviors. Our results indicate the Gauss-Seidel variant is superior in terms of delivering predictable convergence behaviour and hence should be preferred over Jacobi variants. Our results also suggest that Global r -Factor strategies are better for structured stacking scenarios and can achieve absolute convergence in more cases.
Production-Level Facial Performance Capture Using Deep Convolutional Neural Networks
(ACM, 2017) Laine, Samuli; Karras, Tero; Aila, Timo; Herva, Antti; Saito, Shunsuke; Yu, Ronald; Li, Hao; Lehtinen, Jaakko; Bernhard Thomaszewski and KangKang Yin and Rahul Narain
We present a real-time deep learning framework for video-based facial performance capture-the dense 3D tracking of an actor's face given a monocular video. Our pipeline begins with accurately capturing a subject using a high-end production facial capture pipeline based on multi-view stereo tracking and artist-enhanced animations. With 5-10 minutes of captured footage, we train a convolutional neural network to produce high-quality output, including self-occluded regions, from a monocular video sequence of that subject. Since this 3D facial performance capture is fully automated, our system can drastically reduce the amount of labor involved in the development of modern narrative-driven video games or films involving realistic digital doubles of actors and potentially hours of animated dialogue per character.We compare our results with several state-of-the-art monocular real-time facial capture techniques and demonstrate compelling animation inference in challenging areas such as eyes and lips.
Fully Asynchronous SPH Simulation
(ACM, 2017) Reinhardt, Stefan; Huber, Markus; Eberhardt, Bernhard; Weiskopf, Daniel; Bernhard Thomaszewski and KangKang Yin and Rahul Narain
We present a novel method for fully asynchronous time integration of particle-based fluids using smoothed particle hydrodynamics (SPH). With our approach, we allow a dedicated time step for each particle. Therefore, we are able to increase the e ciency of simulations. Previous approaches of locally adaptive time steps have shown promising results in the form of increased time steps, however, they need to synchronize time steps in recurring intervals, which involves either interpolation operations or matching time steps. With our method, time steps are asynchronous through the whole simulation and no global time barriers are needed. In addition, we present an e cient method for parallelization of our novel asynchronous time integration. For both serial and parallel execution, we achieve speedups of up to 7:5 compared to fixed time steps and are able to outperform previous adaptive approaches considerably
Physically-Based Droplet Interaction
(ACM, 2017) Jones, Richard; Southern, Richard; Bernhard Thomaszewski and KangKang Yin and Rahul Narain
In this paper we present a physically-based model for simulating realistic interactions between liquid droplets in an e cient manner. Our particle-based system recreates the coalescence, separation and fragmentation interactions that occur between colliding liquid droplets and allows systems of droplets to be meaningfully represented by an equivalent number of simulated particles. By considering the interactions speci c to liquid droplet phenomena directly, we display novel levels of detail that cannot be captured using other interaction models at a similar scale. Our work combines experimentally validated components, originating in engineering, with a collection of novel modi cations to create a particle-based interaction model for use in the development of mid-to-large scale dropletbased liquid spray e ects. We demonstrate this model, alongside a size-dependent drag force, as an extension to a commonly-used ballistic particle system and show how the introduction of these interactions improves the quality and variety of results possible in recreating liquid droplets and sprays, even using these otherwise simple systems.
Modeling and Data-Driven Parameter Estimation for Woven Fabrics
(ACM, 2017) Clyde, David; Teran, Joseph; Tamstorf, Rasmus; Bernhard Thomaszewski and KangKang Yin and Rahul Narain
Accurate estimation of mechanical parameters for simulation of woven fabrics is essential in many fields. To facilitate this we first present a new orthotropic hyperelastic constitutive model for woven fabrics. Next, we design an experimental protocol for characterizing real fabrics based on commercially available tests. Finally, we create a method for accurately fitting the material parameters to the experimental data. The last step is accomplished by solving inverse problems based on a Catmull-Clark subdivision finite element discretization of the Kirchhoff-Love equations for thin shells. Using this approach we are able to reproduce the fully nonlinear behavior corresponding to the captured data with a small number of parameters while maintaining all fundamental invariants from continuum mechanics. The resulting constitutive model can be used with any discretization (e.g., simple triangle meshes) and not just subdivision finite elements. We illustrate the entire process with results for five types of fabric and compare photo reference of the real fabrics to the simulated equivalents.
Density Maps for Improved SPH Boundary Handling
(ACM, 2017) Koschier, Dan; Bender, Jan; Bernhard Thomaszewski and KangKang Yin and Rahul Narain
In this paper, we present the novel concept of density maps for robust handling of static and rigid dynamic boundaries in fluid simulations based on Smoothed Particle Hydrodynamics (SPH). In contrast to the vast majority of existing approaches, we use an implicit discretization for a continuous extension of the density field throughout solid boundaries. Using the novel representation we enhance accuracy and efficiency of density and density gradient evaluations in boundary regions by computationally efficient lookups into our density maps. The map is generated in a preprocessing step and discretizes the density contribution in the boundary's near-field. In consequence of the high regularity of the continuous boundary density field, we use cubic Lagrange polynomials on a narrow-band structure of a regular grid for discretization. This strategy not only removes the necessity to sample boundary surfaces with particles but also decouples the particle size from the number of sample points required to represent the boundary. Moreover, it solves the ever-present problem of particle deficiencies near the boundary. In several comparisons we show that the representation is more accurate than particle samplings, especially for smooth curved boundaries. We further demonstrate that our approach robustly handles scenarios with highly complex boundaries and even outperforms one of the most recent sampling based techniques.
Emotion Control of Unstructured Dance Movements
(ACM, 2017) Aristidou, Andreas; Zeng, Qiong; Stavrakis, Efstathios; Yin, KangKang; Cohen-Or, Daniel; Chrysanthou, Yiorgos; Chen, Baoquan; Bernhard Thomaszewski and KangKang Yin and Rahul Narain
Motion capture technology has enabled the acquisition of high quality human motions for animating digital characters with extremely high fidelity. However, despite all the advances in motion editing and synthesis, it remains an open problem to modify pre-captured motions that are highly expressive, such as contemporary dances, for stylization and emotionalization. In this work, we present a novel approach for stylizing such motions by using emotion coordinates de ned by the Russell's Circumplex Model (RCM).We extract and analyze a large set of body and motion features, based on the Laban Movement Analysis (LMA), and choose the e ective and consistent features for characterizing emotions of motions. These features provide a mechanism not only for deriving the emotion coordinates of a newly input motion, but also for stylizing the motion to express a di erent emotion without having to reference the training data. Such decoupling of the training data and new input motions eliminates the necessity of manual processing and motion registration. We implement the two-way mapping between the motion features and emotion coordinates through Radial Basis Function (RBF) regression and interpolation, which can stylize freestyle highly dynamic dance movements at interactive rates. Our results and user studies demonstrate the e ectiveness of the stylization framework with a variety of dance movements exhibiting a diverse set of emotions.
Learning Locomotion Skills Using DeepRL: Does the Choice of Action Space Matter?
(ACM, 2017) Peng, Xue Bin; Panne, Michiel van de; Bernhard Thomaszewski and KangKang Yin and Rahul Narain
The use of deep reinforcement learning allows for high-dimensional state descriptors, but little is known about how the choice of action representation impacts learning and the resulting performance. We compare the impact of four di erent action parameterizations (torques, muscle-activations, target joint angles, and target jointangle velocities) in terms of learning time, policy robustness, motion quality, and policy query rates. Our results are evaluated on a gaitcycle imitation task for multiple planar articulated figures and multiple gaits. We demonstrate that the local feedback provided by higher-level action parameterizations can signi cantly impact the learning, robustness, and motion quality of the resulting policies.
Inequality Cloth
(ACM, 2017) Jin, Ning; Lu, Wenlong; Geng, Zhenglin; Fedkiw, Ronald P.; Bernhard Thomaszewski and KangKang Yin and Rahul Narain
As has been noted and discussed by various authors, numerical simulations of deformable bodies often adversely suffer from so-called ''locking'' artifacts. We illustrate that the ''locking'' of out-of-plane bending motion that results from even an edge-spring-only cloth simulation can be quite severe, noting that the typical remedy of softening the elastic model leads to an unwanted rubbery look. We demonstrate that this ''locking'' is due to the well-accepted notion that edge springs in the cloth mesh should preserve their lengths, and instead propose an inequality constraint that stops edges from stretching while allowing for edge compression as a surrogate for bending. Notably, this also allows for the capturing of bending modes at scales smaller than those which could typically be represented by the mesh. Various authors have recently begun to explore optimization frameworks for deformable body simulation, which is particularly germane to our inequality cloth framework. After exploring such approaches, we choose a particular approach and illustrate its feasibility in a number of scenarios including contact, collision, and self-collision. Our results demonstrate the efficacy of the inequality approach when it comes to folding, bending, and wrinkling, especially on coarser meshes, thus opening up a plethora of interesting possibilities.
Long Range Constraints for Rigid Body Simulations
(ACM, 2017) Müller, Matthias; Chentanez, Nuttapong; Macklin, Miles; Jeschke, Stefan; Bernhard Thomaszewski and KangKang Yin and Rahul Narain
The two main constraints used in rigid body simulations are contacts and joints. Both constrain the motion of a small number of bodies in close proximity. However, it is often the case that a series of constraints restrict the motion of objects over longer distances such as the contacts in a large pile or the joints in a chain of rigid bodies. When only short range constraints are considered, a large number of solver iterations is typically needed for long range effects to emerge because information has to be propagated through individual joints and contacts. Our basic idea to signi cantly speed up this process is to analyze the contact or joint graphs and automatically derive long range constraints such as upper and lower distance bounds between bodies that can potentially be far apart both spatially and topologically. The long range constraints are either generated or updated at every time step in case of contacts or whenever their topology changes within a joint graph. The signi cant increase of the convergence rate due to the use of long range constraints allows us to simulate scenarios that cannot be handled by traditional solvers with a number of solver iterations that allow real time simulation.
Augmenting Sampling Based Controllers with Machine Learning
(ACM, 2017) Rajamäki, Joose; Hämäläinen, Perttu; Bernhard Thomaszewski and KangKang Yin and Rahul Narain
E cient learning of 3D character control still remains an open problem despite of the remarkable recent advances in the field. We propose a new algorithm that combines planning by a samplingbased model-predictive controller and learning from the planned control, which is very noisy. We combine two methods of learning: 1) immediate but imprecise nearest-neighbor learning, and 2) slower but more precise neural network learning. The nearest neighbor learning allows to rapidly latch on to new experiences whilst the neural network learns more gradually and develops a stable representation of the data. Our experiments indicate that the learners augment each other, and allow rapid discovery and re nement of complex skills such as 3D bipedal locomotion. We demonstrate this in locomotion of 1-, 2- and 4-legged 3D characters under disturbances such as heavy projectile hits and abruptly changing target direction. When augmented with the learners, the sampling based model predictive controller can produce these stable gaits in under a minute on a 4-core CPU. During training the system runs real-time or at interactive frame rates depending on the character complexity.
Designing Cable-Driven Actuation Networks for Kinematic Chains and Trees
(ACM, 2017) Megaro, Vittorio; Knoop, Espen; Spielberg, Andrew; Levin, David I.W.; Matusik, Wojciech; Gross, Markus; Thomaszewski, Bernhard; Bächer, Moritz; Bernhard Thomaszewski and KangKang Yin and Rahul Narain
In this paper we present an optimization-based approach for the design of cable-driven kinematic chains and trees. Our system takes as input a hierarchical assembly consisting of rigid links jointed together with hinges. The user also specifies a set of target poses or keyframes using inverse kinematics. Our approach places torsional springs at the joints and computes a cable network that allows us to reproduce the specified target poses. We start with a large set of cables that have randomly chosen routing points and we gradually remove the redundancy. Then we refine the routing points taking into account the path between poses or keyframes in order to further reduce the number of cables and minimize required control forces. We propose a reduced coordinate formulation that links control forces to joint angles and routing points, enabling the co-optimization of a cable network together with the required actuation forces. We demonstrate the efficacy of our technique by designing and fabricating a cable-driven, animated character, an animatronic hand, and a specialized gripper.
Authoring Motion Cycles
(ACM, 2017) Ciccone, Loïc; Guay, Martin; Nitti, Maurizio; Sumner, Robert W.; Bernhard Thomaszewski and KangKang Yin and Rahul Narain
Motion cycles play an important role in animation production and game development. However, creating motion cycles relies on general-purpose animation packages with complex interfaces that require expert training. Our work explores the speci c challenges of motion cycle authoring and provides a system simple enough for novice animators while maintaining the flexibility of control demanded by experts. Due to their cyclic nature, we show that performance animation provides a natural interface for motion cycle speci cation. Our system allows the user to act several loops of motion using a variety of capture devices and automatically extracts a looping cycle from this potentially noisy input. Motion cycles for di erent character components can be authored in a layered fashion, or our method supports cycle extraction from higher-dimensional data for capture devices that deliver many degrees of freedom. After capture, a custom curve representation and manipulation tool allows the user to coordinate and control spatial and temporal transformations from a single viewport. Ground and other planar contacts are speci ed with a single sketched line that adjusts a curve's position and timing to establish non-slipping contact. We evaluate the e ectiveness of our work through tests with both novice and expert users and show a variety of animated motion cycles created with our system.
A Micropolar Material Model for Turbulent SPH Fluids
(ACM, 2017) Bender, Jan; Koschier, Dan; Kugelstadt, Tassilo; Weiler, Marcel; Bernhard Thomaszewski and KangKang Yin and Rahul Narain
In this paper we introduce a novel micropolar material model for the simulation of turbulent inviscid fluids. The governing equations are solved by using the concept of Smoothed Particle Hydrodynamics (SPH). As already investigated in previous works, SPH fluid simulations su er from numerical di usion which leads to a lower vorticity, a loss in turbulent details and finally in less realistic results. To solve this problem we propose a micropolar fluid model. The micropolar fluid model is a generalization of the classical Navier- Stokes equations, which are typically used in computer graphics to simulate fluids. In contrast to the classical Navier-Stokes model, micropolar fluids have a microstructure and therefore consider the rotational motion of fluid particles. In addition to the linear velocity field these fluids also have a field of microrotation which represents existing vortices and provides a source for new ones. However, classical micropolar materials are viscous and the translational and the rotational motion are coupled in a dissipative way. Since our goal is to simulate turbulent fluids, we introduce a novel modi ed micropolar material for inviscid fluids with a non-dissipative coupling Our model can generate realistic turbulences, is linear and angular momentum conserving, can be easily integrated in existing SPH simulation methods and its computational overhead is negligible.
Evaporation and Condensation of SPH-based Fluids
(ACM, 2017) Hochstetter, Hendrik; Kolb, Andreas; Bernhard Thomaszewski and KangKang Yin and Rahul Narain
In this paper we present a method to simulate evaporation and condensation of liquids. Therefore, both the air and liquid phases have to be simulated. We use, as a carrier of vapor, a coarse grid for the air phase and mass-preservingly couple it to an SPH-based liquid and rigid body simulation. Since condensation only takes place on rigid surfaces, it is captured using textures that carry water to achieve high surface detail. The textures can exchange water with the air phase and are used to generate new particles due to condensation effects yielding a full two-way coupling of air phase and liquid. In order to allow gradual evaporation and condensation processes, liquid particles can take on variable sizes. Our proposed improved implicit surface definition is able to render dynamic contact angles for moving droplets yielding highly detailed fluid rendering.
A Positive-Definite Cut-Cell Method for Strong Two-Way Coupling Between Fluids and Deformable Bodies
(ACM, 2017) Zarifi, Omar; Batty, Christopher; Bernhard Thomaszewski and KangKang Yin and Rahul Narain
We present a new approach to simulation of two-way coupling between inviscid free surface fluids and deformable bodies that exhibits several notable advantages over previous techniques. By fully incorporating the dynamics of the solid into pressure projection, we simultaneously handle fluid incompressibility and solid elasticity and damping. Thanks to this strong coupling, our method does not su er from instability, even in very taxing scenarios. Furthermore, use of a cut-cell discretization methodology allows us to accurately apply proper free-slip boundary conditions at the exact solid-fluid interface. Consequently, our method is capable of correctly simulating inviscid tangential flow, devoid of grid artefacts or arti cial sticking. Lastly, we present an e cient algebraic transformation to convert the inde nite coupled pressure projection system into a positive-de nite form.We demonstrate the e cacy of our proposed method by simulating several interesting scenarios, including a light bath toy colliding with a collapsing column of water, liquid being dropped onto a deformable platform, and a partially liquid- lled deformable elastic sphere bouncing.
Hierarchical Vorticity Skeletons
(ACM, 2017) Eberhardt, Sebastian; Weissmann, Steffen; Pinkall, Ulrich; Thuerey, Nils; Bernhard Thomaszewski and KangKang Yin and Rahul Narain
We propose a novel method to extract hierarchies of vortex filaments from given three-dimensional flow velocity fields. We call these collections of filaments Hierarchical Vorticity Skeletons (HVS). They extract multi-scale information from the input velocity field, which is not possible with any previous filament extraction approach. Once computed, these HVSs provide a powerful mechanism for data compression and a very natural way for modifying flows. The data compression rates for all presented examples are above 99%. Employing our skeletons for flow modification has several advantages over traditional approaches. Most importantly, they reduce the complexity of three-dimensional fields to one-dimensional lines and, make complex fluid data more accessible for changing de ning features of a flow. The strongly reduced HVS dataset still carries the main characteristics of the flow. Through the hierarchy we can capture the main features of di erent scales in the flow and by that provide a level of detail control. In contrast to previous work, we present a fully automated pipeline to robustly decompose dense velocities into filaments.

Browse

Browsing SCA 17: Eurographics/SIGGRAPH Symposium on Computer Animation by Issue Date

Results Per Page

Sort Options