Frédéric H. Pighin

Motion graphs

In this paper we present a novel method for creating realistic, controllable motion. Given a corpus of motion capture data, we automatically construct a directed graph called a motion graph that encapsulates connections among the database. The motion graph consists both of pieces of original motion and automatically generated transitions. Motion can be generated simply by building walks on the graph. We present a general framework for extracting particular graph walks that meet a users specifications. We then show how this framework can be applied to the specific problem of generating different styles of locomotion along arbitrary paths.

Expressive speech-driven facial animation

Speech-driven facial motion synthesis is a well explored research topic. However, little has been done to model expressive visual behavior during speech. We address this issue using a machine learning approach that relies on a database of speech-related high-fidelity facial motions. From this training set, we derive a generative model of expressive facial motion that incorporates emotion control, while maintaining accurate lip-synching. The emotional content of the input speech can be manually specified by the user or automatically extracted from the audio signal using a Support Vector Machine classifier.

Hybrid control for interactive character animation

We implement a framework for animating interactive characters by combining kinematic animation with physical simulation. The combination of animation techniques allows the characters to exploit the advantages of each technique. For example, characters can perform natural-looking kinematic gaits and react dynamically to unexpected situations. Kinematic techniques such as those based on motion capture data can create very natural-looking animation. However, motion capture based techniques are not suitable for modeling the complex interactions between dynamically interacting characters. Physical simulation, on the other hand, is well suited for such tasks. Our work develops kinematic and dynamic controllers and transition methods between the two control methods for interactive character animation. In addition, we utilize the motion graph technique to develop complex kinematic animation from shorter motion clips as a method of kinematic control.

Unsupervised learning for speech motion editing

We present a new method for editing speech related facial motions. Our method uses an unsupervised learning technique, Independent Component Analysis (ICA), to extract a set of meaningful parameters without any annotation of the data. With ICA, we are able to solve a blind source separation problem and describe the original data as a linear combination of two sources. One source captures content (speech) and the other captures style (emotion). By manipulating the independent components we can edit the motions in intuitive ways.

Modeling and editing flows using advected radial basis functions

Fluid simulations are notoriously difficult to predict and control. As a result, authoring fluid flows often involves a tedious trial and error process. There is to date no convenient way of editing a fluid after it has been simulated. In particular, the Eulerian approach to fluid simulation is not suitable for flow editing since it does not provide a convenient spatio-temporal parameterization of the simulated flows. In this research, we develop a new technique to learn such parameterization. This technique is based on a new representation, the <i>Advected Radial Basis Function</i>. It is a time-varying kernel that models the local properties of the fluid. We describe this representation and demonstrate its use for interactive three-dimensional flow editing.

Real-time speech motion synthesis from recorded motions

Data-driven approaches have been successfully used for realistic visual speech synthesis. However, little effort has been devoted to real-time lip-synching for interactive applications. In particular, algorithms that are based on a graph of motions are notorious for their exponential complexity. In this paper, we present a greedy graph search algorithm that yields vastly superior performance and allows real-time motion synthesis from a large database of motions. The time complexity of the algorithm is linear with respect to the size of an input utterance. In our experiments, the synthesis time for an input sentence of average length is under a second.

Using multiagent teams to improve the training of incident commanders

The DEFACTO system is a multiagent based tool for training incident commanders for large scale disasters. In this paper, we highlight some of the lessons that we have learned from our interaction with the Los Angeles Fire Department (LAFD) and how they have affected the way that we continued the design of our training system. These lessons were gleaned from LAFD feedback and initial training exercises and they include: system design, visualization, improving trainee situational awareness, adjusting training level of difficulty and situation scale. We have taken these lessons and used them to improve the DEFACTO systems training capabilities. We have conducted initial training exercises to illustrate the utility of the system in terms of providing useful feedback to the trainee.

Extended Galilean invariance for adaptive fluid simulation

In an unbounded physical domain, simulating a turbulent fluid on an Eulerian grid is rather tricky. Since it is difficult to predict the motion of the fluid, it is also difficult to guess which computational domain would allow the simulation of the fluid without crossing the computational boundaries. To address this dilemma, we have developed a novel adaptive framework where the simulation grid follows the motion of the flow. Our technique is based on the principle of Galilean Invariance and the culling of simulation cells using a metric derived from continuative boundary conditions. We describe our framework and showcase its advantages over traditional techniques. Timing results and visual comparisons are presented.

Scattered data interpolation for computer graphics

The goal of scattered data interpolation techniques is to construct a (typically smooth) function from a set of unorganized samples. These techniques have a wide range of applications in computer graphics and computer vision. For instance they can be used to model a surface from a set of sparse samples, to reconstruct a BRDF from a set of measurements, or to interpolate motion capture data. This course will survey and compare scattered interpolation algorithms and describe their applications in computer graphics. Although the course is focused on applying these techniques, we will introduce some of the underlying mathematical theory and briefly mention numerical considerations.

Fluid simulation via disjoint translating grids

Most computer graphics applications have leaned towards the Eulerian approach, where the Navier-Stokes equations are expressed in a fixed global coordinate frame. Finite-difference methods can be made unconditionally stable with semi-Lagrangian methods as suggested by [Stam 1999], and have been used to simulate gases and liquids. Eulerian methods excel at representing interfaces via Levelset methods, and incompressibility can be enforced efficiently using the projection method. While Eulerian methods can be unconditionally stable, excessively large time steps introduce severe numerical viscosity, and hence ruin physical and visual accuracy. This is why [Enright et al. 2002] recommends limiting the time step to no more than 5 times the Courant-Friedrich-Levy (CFL) condition. For splashy simulations, this can lead to excessively small time steps.