François Faure

Grimage: markerless 3D interactions

Grimage glues multi-camera 3D modeling, physical simulation and parallel execution for a new immersive experience. Put your hands or any object into the interaction space. It is instantaneously modeled in 3D and injected into a virtual world populated with solid and soft objects. Push them, catch them and squeeze them.

A layered model of a virtual human intestine for surgery simulation.

In this paper, we propose a new approach to simulate the small intestine in a context of laparoscopic surgery. The ultimate aim of this work is to simulate the training of a basic surgical gesture in real-time: moving aside the intestine to reach hidden areas of the abdomen. The main problem posed by this kind of simulation is animating the intestine. The problem comes from the nature of the intestine: a very long tube which is not isotropically elastic, and is contained in a volume that is small when compared to the intestines length. It coils extensively and collides with itself in many places. To do this, we use a layered model to animate the intestine. The intestines axis is animated as a linear mechanical component. A specific sphere-based model handles contacts and self-collisions. A skinning model is used to create the intestines volume around the axis. This paper discusses and compares three different representations for skinning the intestine: a parametric surface model and two implicit surface models. The first implicit surface model uses point skeletons while the second uses local convolution surfaces. Using these models, we obtained good-looking results in real-time. Some videos of this work can be found in the online version at doi: 10.1016/j.media.2004.11.006 and at www-imagis.imag.fr/Publications/2004/FLAMCFC04.

Implicit FEM Solver on GPU for Interactive Deformation Simulation

Publisher Summary This chapter presents a set of methods to implement an implicit Finite Element solver on the graphics processing units (GPU). The Finite Element Method (FEM) is broadly used to simulate deformable materials in physics simulations. Existing GPU-based methods implement FEM only with explicit time integrators, which are simple and easy to parallelize. However, these methods suffer from stability issues and require very small time steps to simulate stiff materials. The finite element method provides a means for discretizing and solving volumetric models of deformable materials. To parallelize a given set of computations on the GPU it is necessary to extract a massive level of parallelism, on the order of tens of thousands of threads. In many cases, the computations are independent except that they need to scatter their results onto a set of shared variables. This happens for instance when computing FEM elements accumulating forces to, or when computing a sparse matrix–vector product. Such operations can be represented as a graph, where nodes represent shared variables and edges the computations between them. A very common technique to parallelize such a graph is to partition it into a set of subgraphs, each computed by a different processor. To achieve maximum performance it is critical to design data layouts to optimize memory access. A common approach to improve cache efficiency of memory access patterns is to convert large arrays of structures into a structure of arrays. The fastest method to compute the local frame of an element, although not the most accurate one, is to set the origin at vertex p0 and use its adjacent edges.

Animating Shapes at Arbitrary Resolution with Non-Uniform Stiffness

We present a new method for physically animating deformable shapes using finite element models (FEM). Contrary to commonly used methods based on tetrahedra, our finite elements are the bounding voxels of a given shape at arbitrary resolution. This alleviates the complexities and limitations of tetrahedral volume meshing and results in regular, well-conditionned meshes. We show how to build the voxels and how to set the masses and stiffnesses in order to model the physical properties as accurately as possible at any given resolution. Additionally, we extend a fast and robust tetrahedron-FEM approach to the case of hexahedral elements. This permits simulation of arbitrarily complex shapes at interactive rates in a manner that takes into account the distribution of material within the elements.

DEM-based simulation of concrete structures on GPU

The benefit of using the discrete element method (DEM) for simulations of fracture in heterogeneous media has been widely highlighted. However, modelling large structure leads to prohibitive computations times. We propose to use graphics processing units (GPUs) to reduce the computation time, taking advantage of the highly data parallel nature of DEM computations. GPUs are massively parallel coprocessors increasingly popular to accelerate numerical simulations. We detail our algorithm and implementation of the DEM on GPU and present performance results for simulations of rock impact on a concrete slab, before to discuss the pro and cons of moving such computation to the GPU. Pour simuler des structures soumises à de la fracturation, la méthode des éléments discrets (DEM) constitue un outil très efficace. Cependant la modélisation de grandes structures est très coûteuse en opération. Suite à lobservation que ces calculs sont fortement data-parallèles, nous proposons de tirer partie des processeurs graphiques (GPUs) pour réduire le temps de calcul. Les GPUs sont des coprocesseurs massivement parallèles de plus en plus utilisés pour accélérer des simulations numériques. Lalgorithme et limplantation sur le GPU sont détaillés puis nous présentons les résultats obtenus pour une simulation dimpact sur dalle en béton.

Exploring the Use of Adaptively Restrained Particles for Graphics Simulations

In this paper, we explore the use of Adaptively Restrained (AR) particles for graphics simulations. Contrary to previous methods, Adaptively Restrained Particle Simulations (ARPS) do not adapt time or space sampling, but rather switch the positional degrees of freedom of particles on and off, while letting their momenta evolve. Therefore, inter-particles forces do not have to be updated at each time step, in contrast with traditional methods that spend a lot of time there. We present the initial formulation of ARPS that was introduced for molecular dynamics simulations, and explore its potential for Computer Graphics applications: We first adapt ARPS to particle-based fluid simulations and propose an efficient incremental algorithm to update forces and scalar fields. We then introduce a new implicit integration scheme enabling to use ARPS for cloth simulation as well. Our experiments show that this new, simple strategy for adaptive simulations can provide significant speedups more easily than traditional adaptive models.