Current technology allows for many interesting and exciting scenarios in games. Including larger crowds, maybe in a city or a group of soldiers on a battlefield, could provide greater immersion for players if the flow of the crowd can be accurately predicted. Stephen J. Guy and Ioannis Karamouzas of the University of Minnesota and Brian Skinner from Argonne National Laboratory have been researching real-time crowd simulations with some interesting results. They recently published a paper in Physical Review Letters focusing on their data analysis, accomplished with an Intel laptop we provided them. The analysis revealed a universal power law that governs pedestrian behavior can predict and simulate crowd flows.
More details, including the paper, C++ and python code samples are available here: http://motion.cs.umn.edu/PowerLaw/