Level of Detail Techniques for Accelerating Crowd Simulations

In this work we apply level of detail techniques to scale up the number of agents in realtime crowd simulations. Level of detail techniques are most studied with respect to geometric simplifications for rendering, but similar ideas are used here for computational simplification for simulations. In particular, we studied methods for computing the paths of agents in a crowd. The main approach for simplification is to cluster nearby agents in a similar state into one larger virtual agent, then simulate that virtual agent with an existing crowd simulation. We may then apply the changes in the virtual agent to the real agents in the corresponding cluster. We measure how much speed this technique gains in three different simulators and highlight cases where this approach is significantly faster than previously accepted methods.