ERC Starting Grant. Duration 2012-2016.
Real-world mechanical effects exhibit multiple sources of complexity, including nonlinearity, anisotropy, and heterogeneity. Due to this complexity, simulation methods in computer animation must face a trade-off between realism and efficiency. The classical way to achieve highly realistic simulations of mechanical effects is to use complex constitutive models, combined with tedious parameterization.
In the Animetrics project, instead, we investigate novel models for physically based simulation in computer animation, by exploiting data from example effects. We propose to fit simple models in local operations domains, and then combine the models in nonlinear ways to achieve the desired global behavior. The example data, either captured from real-world effects, or precomputed in accurate simulations, serves to estimate the parameters of the local models and to design the interpolation procedures.
Yarn-level simulation of woven cloth
The large-scale mechanical behavior of woven cloth is determined by the mechanical properties of the yarns, the weave pattern, and frictional contact between yarns. Using standard simulation methods for elastic rod models and yarn-yarn contact handling, the simulation of woven garments at realistic yarn densities is deemed intractable. We introduce an efficient solution for simulating woven cloth at the yarn level. Central to our solution is a novel discretization of interlaced yarns based on yarn crossings and yarn sliding, which allows modeling yarn-yarn contact implicitly, avoiding contact handling at yarn crossings altogether. Combined with models for internal yarn forces and inter-yarn frictional contact, as well as a massively parallel solver, we are able to simulate garments with hundreds of thousands of yarn crossings at practical framerates on a desktop machine, showing combinations of large-scale and fine-scale effects induced by yarn-level mechanics (Cirio et al. 2014).
Interactive deformation of heterogeneous volume data
In (Torres et al. 2014) we present a method to interactively deform volume images with heterogeneous structural content, using coarse tetrahedral meshes. It rests on two major components: a massively parallel algorithm for the rasterization of tetrahedral meshes, and a method to define a coarse deformable tetrahedral mesh from the homogenization of a fine heterogeneous mesh. We show the potential of the method for training and planning applications through two examples: an abdominal CT exploration and the alignment of breast CT and MRIs.
Physics-aware Voronoi fracture
In (Schvartzman and Otaduy 2014a & 2014b) we present a novel algorithm to simulate brittle fracture. It augments previous methods based on Voronoi diagrams, improving their versatility and their ability to adapt fracture patterns automatically to diverse collision scenarios and object properties. We cast brittle fracture as the computation of a high-dimensional Centroidal Voronoi Diagram (CVD), where the distribution of fracture fragments is guided by the deformation field of the fractured object. By formulating the problem in high dimensions, we support robustly object and crack concavities, as well as intuitive artist control. We further accelerate the fracture animation process with example-based learning of the fracture degree, and a highly parallel tessellation algorithm. As a result, we obtain fast animations of detailed and rich fractures, with fracture patterns that adapt to each particular collision scenario.
Simulating articulated subspace self-contact (in colaboration with UC Santa Barbara)
We present an efficient new subspace method for simulating the self-contact of articulated deformable bodies, such as characters. Self-contact is highly structured in this setting, as the limited space of possible articulations produces a predictable set of coherent collisions. Subspace methods can leverage this coherence, and have been used in the past to accelerate the collision detection stage of contact simulation. We show that these methods can be used to accelerate the entire contact computation, and allow self-contact to be resolved without looking at all of the contact points. Our analysis of the problem yields a broader insight into the types of non-linearities that subspace methods can efficiently approximate, and leads us to design a pose-space cubature scheme. Our algorithm accelerates self-contact by up to an order of magnitude over other subspace simulations, and accelerates the overall simulation by two orders of magnitude over full-rank simulations. We demonstrate the simulation of high resolution (100K – 400K elements) meshes in self-contact at interactive rates (5.8 – 50 FPS) (Teng et al. 2014).
Strain limiting for soft-finger contact simulation
The command of haptic devices for rendering direct interaction with the hand requires thorough knowledge of the forces and deformations caused by contact interactions on the fingers. In (Perez et al. 2013), we propose an algorithm to simulate nonlinear elasticity under frictional contact, with the goal of establishing a model-based strategy to command haptic devices and to render direct hand interaction. The key novelty in our algorithm is an approach to model the extremely nonlinear elasticity of finger skin and flesh using strain-limiting constraints, which are seamlessly combined with frictional contact constraints in a standard constrained dynamics solver. We show that our approach enables haptic rendering of rich and compelling deformations of the fingertip.
Modeling and estimation of internal friction in cloth
Force-deformation measurements of cloth exhibit significant hysteresis, and many researchers have identified internal friction as the source of this effect. However, it has not been incorporated into computer animation models of cloth. In (Miguel et al. 2013), we propose a model of internal friction based on an augmented reparameterization of Dahl’s model, and we show that this model provides a good match to several important features of cloth hysteresis even with a minimal set of parameters. We also propose novel parameter estimation procedures that are based on simple and inexpensive setups and need only sparse data, as opposed to the complex hardware and dense data acquisition of previous methods. Finally, we provide an algorithm for the efficient simulation of internal friction, and we demonstrate it on simulation examples that show disparate behavior with and without internal friction.
Data-driven estimation of cloth models
Progress in cloth simulation for computer animation and apparel design has led to a multitude of deformation models, each with its own way of relating geometry, deformation, and forces. As simulators improve, differences between these models become more important, but it is difficult to choose a model and a set of parameters to match a given real material simply by looking at simulation results. We have designed novel methods to measure force-deformation examples on sample fabrics, and to fit nonlinear mechanical models to the observed deformations. Unlike standard textile testing, our approach measures complex 3D deformations of a sheet of cloth, not just one-dimensional force-displacement curves, so it works under a wider range of deformation conditions. Our nonlinear mechanical models are expressed as the interpolation of simpler linear models. We have evaluated the fitted models by comparison to measured deformations with motions very different from those used for fitting (Miguel et al. 2012).
Example-based animation of cloth wrinkles
The simulation of cloth with rich folds and wrinkles is a computationally expensive process. To circumvent this cost, we have designed an example-based algorithm for fast animation of plausible cloth wrinkles. Unlike other approaches, our algorithm does not depend on a character’s pose, therefore it is valid for loose dresses, curtains, etc., not just cloth defined by skinning techniques. Central to our approach is a correspondence between low- and high-resolution cloth deformations, both at the training and synthesis stages. Based on this correspondence, we define an algorithm for synthesizing cloth wrinkles as a function of the deformation of a low-resolution cloth and a set of example poses. We have demonstrated the animation of plausible high-resolution wrinkles at high frame rates, suitable for interactive applications such as video games (Zurdo et al. 2012).
One category of data-driven methods builds on traditional mechanics models, but augments the classic constitutive models with data from example effects. In particular, we have designed data-driven methods to simulate in an efficient and realistic manner nonlinear heterogeneous elastic solids. Our solution defines the stress-strain relationship of the material using an interpolation of local linear stress-strain models. These models are also spatially local in order to capture spatial variation of properties. The parameters of the complete object are fit simultaneously by solving an inverse deformation problem based on samples of force-deformation fields (Bickel et al. 2009). Interestingly, we have applied such example-based materials to design computational methods for fabrication of heterogeneous solids (Bickel et al. 2010). We have also exploited example-based materials to design artist-centric animation control methods, which allow pose-dependent dynamic behavior (Galoppo et al. 2009).
Another category of data-driven methods models directly the shape of the animated objects by combining data from precomputed examples. In particular, we have designed a method for the animation of facial wrinkles from a few example expressions. Our method decomposes facial deformation into a large-scale (i.e., expression) and a small-scale (i.e., wrinkles). Given this decomposition, the small-scale deformation can be accurately approximated in the low-dimensional domain of large-scale deformations, through local combination of example wrinkles (Bickel et al. 2007 & 2008).