Crowd Simulation

A crowd simulation is a computer program that simulates the motion of a human crowd. In the research area of crowd simulation, we aim to develop algorithms that can simulate large crowds realistically and efficiently.

Realistic real-time crowd simulation has many applications. It can help analyze or predict the movement of a crowd in a real-world scenario, such as an evacuation, a concert, or another large-scale event. It can also help enrich games and immersive VR experiences, in which crowds need to respond to the player’s actions.

Our group conducts crowd-simulation research from several angles of approach:

  • doing real-world measurements to understand the walking behavior of humans;
  • designing navigation algorithms to model the behavior of virtual humans;
  • synchronizing between real and virtual crowds;
  • developing theories for analyzing and comparing different navigation algorithms;
  • leveraging machine learning to replicate complex behaviors.

Simulation or animation?

It is important to see the difference between a simulation and an animation. To make a crowd simulation as efficient as possible, we usually simplify each pedestrian to a disk, and we simplify the environment to a 2D plane with obstacles. The output of a simulation is a sequence of trajectories for these disks. If desired, this output can be converted to a video with human-like 3D characters, but this is considered to be a separate step.

Navigation algorithms: Modelling pedestrian motion

Pedestrian navigation algorithms are mathematical expressions that capture the way in which humans walk towards a goal while avoiding (or interacting with) other agents. We can distinguish between the following types of algorithms:

  • Rule-based algorithms are reactive systems that map states to actions: for example, if a collision is imminent, perform an avoidance manoeuvre.
  • Position- and velocity-based algorithms model the interactions between pedestrians as a product of their respective positions and/or velocities.
  • Vision-based algorithms establish a relation between pedestrian motion and what the pedestrian is seeing.

Comments are closed.