INTERESTS

Over the years I've had the opportunity to dabble in many different fields. There are many other fields in which I look forward to playing in the years to come. Because of my experience as a modeler and animator, I tend to gravitate towards those disciplines that contribute to the story-telling process. There is an exception to this: hardware. I have an absolute fascination with hardware design and have had the opportunity to study architecture under Dr. Fred Brooks--a once in a lifetime opportunity.

I'm most interested in subjects in rendering, simulation, and motion synthesis. In the last few years, my time has mostly been consumed by crowds, collision detection, ray tracing, numerical methods and non-photorealistic rendering.

ACTIVE RESEARCH

Crowds

While at UNC, I've had the opportunity to look at the interesting problem of crowd simulation. UNC's work in this regard naturally arose from our heritage in motion planning for multiple, independent agents; simulating a group of humans is rather more complex. My personal interests have tended toward these differences.

• Humans have certain physical constraints in how they can move from point A to point B. Point robots (often used in planning) don't.
• Humans have personality and character which can affect how they value a particular path to a goal. This "value" can be a far more elaborate function than simply calculating the Euclidian distance to goal.
• Humans can interact in complex and subtle ways. Tacit communication often takes place to avoid collisions.
• Planning for the center of mass often leads to huge problems in simulating the actual motion of the individual human agents. I feel the motion constraints (kinematic and dynamic) need to be pushed into the planning stage.

I've had the opportunity to work on a number of papers and to explore some of the higher-level abstractions of human behavior in crowd simulation.

Crowd Visualization

Recently I've been developing UNC's software for visualizing crowds. We've found that planning is good, but if shown in a compelling context, the planning feels more relevant and successful. We've currently chosen the Horde3D engine as our basis for real-time visualization. We've tweaked the engine, tied it into our planner and started developing content. To the left is our subway scenario (currently incomplete -- note no texture on the subway walls.) We hope to simulation realistic crowds in a complex dynamic environment. We also have environments for office evacuation and are developing a stadium evacuation as well. In the images to the left, I modeled the environment, created the textures and made the animation as well.

Our initial visualization relied on a different aesthetic. We created a cartoon world which suggest human-like characters but stays away from the uncanny valley of full human synthesis. The images to the left are samples from our work first at VR07 and then later in an article in TVCG. I created all models, animation, and then lit and rendered the scenes.

Collision Detection

When I first arrived at UNC, I began working in acceleration structures for ray tracing. Many of the acceleration structures common for ray tracing are equally ubiquitous in collision detection. So, moving into collision detection was a natural transition. I worked on Walt Disney Animation Studios' cloth simulator during the summer of 2007. I was given the significant task of taking collision detection time to zero. When I left, I'd tripled the average speed of collision detection and had some ideas that led to two different papers.

While at Disney, I developed the idea of Representative-Triangles. In collaboration with my mentor there, we developed the idea into a full paper which I subsequently presented at I3D2008. Representative-Triangles (R-Tris) seek to improve collision detection performance by improving culling efficiency and eliminating duplicate queries in a simple and efficient manner. The image to the left is one of the benchmarks from that paper. The animation and simulation comes from Walt Disney Animation Studios. The materials, lighting and camera are my own. Alternatively, I also developed an idea called an Orphan Set, which is currently implemented in Disney's system. The details can be found in this paper.

PAST RESEARCH

Ray tracing

I had the opportunity to work with Christian Lauterbach in fall of 2006. Along with Sung-Eui Yoon, we worked on ray tracing large and complex scenes.

My efforts were focused on efficient management of acceleration structures for dynamic scenes. I assisted Sung-Eui Yoon in developing a system called Selective Restructuring which intelligently evaluated the BVH used for ray tracing. The algorithm determines when the hierarchy becomes degraded to the point where a restructuring is justified. More particularly, the algorithm identifies a localized portion of the hierarchy most responsible for the degradation and restructures, as efficiently as possible, only that portion. The images to the left show the benchmarks used in the Selective Restructuring paper.

Numerical Methods

My final year as an undergrad I worked with Mike Kirby at SCI (in collaboration with Jennifer Ryan at Virgina Tech) on a numerical post-processor.

The product of finite-element methods (FEM) is a set of piecewise continuous functions, defined over the domain of simulation. The order of the function on each cell is dependent on the details of the specific FEM techniques used, but at the cell boundaries there are usually not even guarantees of C0 continuity.

Our post-processor filters the FEM results by convolving it with a kernel formed from a linear combination of b-splines. The filtered data guarantees continuity at the boundaries and produces a higher order of accuracy.

Many numerical calculations (such as high-order integration) rely on continuity assumptions. When performing integration, such as creating streamlines, these high-order integrators fail to converge as they should because of the broken continuity assumption. Our filter restores the assumption and allows full convergence. Papers related to this work can be seen here and here.

Non-photorealistic Rendering

In my second year at the University of Utah, I worked with Pete Shirley on developing a real-time non-photorealistic renderer for terrain. The result of my efforts can be seen here. This work went on to form the basis of my undergraduate thesis.