Research

The objective is reducing the time that's spent on dynamic calculations of the moving avatars in a virtual environment. In order to accomplish this objective, we're focusing on creation and maintenance of a minimal "active subset" of the avatars that we're actually interested in. The exact dynamics calculations are performed only for the avatars that are in this active subset. All the rest are approximated by a probability function that's easier to evaluate. This subset is dynamically managed: ie, if an avatar becomes uninteresting (not visible by the viewer), it's pruned from the set and avatars that become interesting are added dynamically. This maintenance step is called "culling". The magic comes in when we try to make sure what viewer sees is not different from the result that we'd get from the exact dynamics simulation of all the avatars in our virtual environments (including those the viwer does not see).

I'll be putting some screenshots of our current system shortly.  

Motion Synthesis & Retargeting:

The objective is to generate motions that look "natural" (i.e. human) and satisfy some user specified constraints. These constraints are used to control the motion so that we generate motions that do what we want. This is important in a production environment where an animator may have strict goals that a motion has to achieve. I.e., a synthetic character may have to interact with a real object. Or, in a game environment, characters may have to behave according to constraints generated by an AI or path planning system. We synthesize the novel motions by cutting pieces from previously generated (key-frames etc.) or captured motions and putting them together. We ensure that the resulting motion is "natural looking" by only using compatible pieces together. We solve the constraint problem by formulating the synthesis as a search problem in the space of available motions that can be obtained from from out motion database. See our paper for details and results of this process.

Motion Synthesis from Annotations:

We may also want to constrain motions in a higher (or lower depending on which way you want to look at it) level. For example, we may want a motion to go to a particular position by walking. If no constraint on the type of motion that should be performed, then the search algorithm can generate a motion that runs very fast and simply waits at the target position. This is rarely what we want. The framework presented in this paper allows animators create a vocabulary to control the motion synthesis in an intuitive way. The vocabulary is freely chosen by the user depending on the features of motion that the user wishes to control.

Animating Suspended Particle Explosions

This paper describes a method for animating suspended particle explosions. Rather than modeling the numerically troublesome, and largely invisible blast wave, the method uses a relatively stable incompressible fluid model to account for the motion of air and hot gases. The fluid's divergence field is adjusted directly to account for detonations and the generation and expansion of gaseous combustion products. Particles immersed in the fluid track the motion of particulate fuel and soot as they are advected by the fluid. Combustion is modeled using a simple but effective process governed by the particle and fluid systems. The method has enough flexibility to also approximate sprays of burning liquids. This paper includes several demonstrative examples showing air bursts, explosions near obstacles, confined explosions, and burning sprays. Because the method is based on components that allow large time integration steps, it only requires a few seconds of computation per frame for the examples shown.

Pixie is a photo-realistic renderer that implement Pixar's RenderMan interface. It's features include raytracing, global illumination with photon mapping and irradiance caching, programmable shading, motion blur, depth of field, texture/environment/shadow mapping, occlusion culling, area light sources, network parallel rendering, DSO shading, REYES style scanline rendering etc. I have been working on this renderer for quite some time, and I'm interested to know if you use my renderer to create nice pictures.

The RenderMan (R) Interface Procedures and RIB Protocol are:
Copyright 1988, 1989, Pixar. All rights reserved.