, University of Pennsylvania
, Trinity College Dublin
Pages: pp. 17-18
Computer graphics' success in creating realistic animated characters has led to a recent upsurge of interest in creating a multiplicity of such human figures engaged in group activities. This expansion of graphical capabilities leads to a wide range of issues: low-level graphical performance and throughput, individual behavior specification and generation, collision avoidance and interaction response, 3D environmental interactions, and overall scenario scripting. The number of papers in the literature relevant to these areas is increasing rapidly; we conceived this special issue as a way to recognize that modeling collective activities has become the central focus for many computer graphics research groups.
We further realized that calling this enterprise "crowd modeling" wouldn't adequately reflect the necessary scope of the expected contributors. So, we adopted the term "virtual populace."
We're pleased to present five papers representing an excellent cross-section of current research and applications in this exciting area. In "Brain Springs: Fast Physics for Large Crowds in WALL•E," Paul Kanyuk provides a behind-the-scenes glimpse of how Pixar simulated crowds of robot and human characters in the animated film. They simulated character physics in a realistic, interactive way, through clever and novel use of Massive, an agent-based crowd system.
Simulating massive crowds in real time necessitates fully exploiting new hardware architectures, as Erdal Yilmaz and his colleagues demonstrate in "The Virtual Marathon: Parallel Computing Supports Crowd Simulations." They explain how they used Nvidia's CUDA (originally Compute Unified Device Architecture) to accelerate the simulation of millions of virtual marathoners, using fuzzy logic.
Sébastien Paris and Stéphane Donikian discuss simulation of more complex human behaviors such as queuing, sourcing information, or making purchases in "Activity-Driven Populace: A Cognitive Approach to Crowd Simulation." In their system, each agent's movements and actions are driven by high-level goals and objectives, which are realized using multilayer cognitive decision-making.
In "YaQ: An Architecture for Real-Time Navigation and Rendering of Varied Crowds," Jonathan Maïm and his colleagues describe a complete system for rendering, animation, and behavioral simulation of large crowds. They also give a good overview of the main components such systems must implement.
Finally, Christopher Peters and Cathy Ennis highlight an important question: How do we ensure that simulations of virtual populace meet a viewer's expectations of how people actually behave? In "Modeling Groups of Plausible Virtual Pedestrians," they present a methodology that uses information extracted from video and a set of perceptual experiments to determine what's most important when simulating small groups of people in a crowded area.
We hope these articles convey some of the scope and excitement of animating a virtual populace. This area challenges us to explore novel agent architectures as well as graphical optimizations and perceptual evaluation. In our own recent research, we've explored some of these challenges, such as higher-level control aspects of functional groups and individuals 1 and exploiting perceptual knowledge to enhance the animation of richly varied crowds of people. 2
Although this overview isn't at all a survey, it would be an oversight not to acknowledge Craig Reynolds' seminal work on flocks and herds. 3 A more recent resource for the interested reader is Daniel Thalmann and Soraia Raupp Musse's book Crowd Simulation. 4
Computer animation's future appears integrally tied to the specification and production of large-scale groups with coordinated, meaningful behaviors. We hope the articles in this special issue help point the way.