To date, crowds in games are usually driven by static pathfinding combined with localized steering and collision avoidance. This technique is well understood and works well for small or moderate crowd sizes that are sparsely distributed in their environment. As crowd density increases and the agents’ goals converge, this approach falls apart. In situations like this, agents fight the rest of the crowd to follow their ideal path as closely as possible, seemingly ignoring all other possible routes.
Congestion maps provide a simple means of modeling aggregate behavior in large crowds of tens or hundreds of thousands of agents. Combined with vector flow ...