Special Konversatorium on Tuesday, March 3, 2015 - 15:45

Recent advances in scanning, modeling, rendering, and hardware make it possible to generate nearphotorealistic images of moderately complex scenes at interactive rates. One of the next grand challenges in computer graphics and visualization is to model vibrant, dynamic scenes of realworld complexity, such as urban spaces. The problem of modeling virtual cityscapes offers a diverse set of opportunities for innovations and provides enabling technologies of societal interests, including energy use, transportation mechanisms, economic sustainability, education and entertainment. Some of the key research issues include interactive simulation of large-scale crowds, realistic modeling of complex traffic flows, efficient motion synthesis of plausible pedestrian behaviors and natural phenomena. At least one to two orders of magnitude performance improvement in hardware will be needed. New algorithms and software systems that can exploit such computing power must be developed.

In this talk, I will survey some of recent efforts on addressing the problem of modeling, simulating, and directing virtual agents in complex dynamic environments. In particular, I will present several complementary approaches for representing movement of multiple virtual entities, including both crowds and traffic, in urban scenes and city highways. I will further highlight the design of scalable algorithms for these problems by taking advantages of parallelism available on emerging commodity hardware, such as GPUs and many-core processors. These methods can be applied to interactive crowd simulation, motion synthesis, and coordination of multiple autonomous agents in computer games and virtual environment systems. I will conclude by discussing our experiences and some future research directions on incorporating sound effects and natural phenomena.