What Causes Deadly 'Crowd-Quakes'?

In July of 2010, more than 1 million people flocked to Duisburg, Germany, for the Love Parade—an annual festival dedicated to techno music. As revelers crammed into a bottlenecked tunnel that served as the event's entrance, their pushing and shoving rippled through the masses, culminating in a “crowd-quake” of violent and chaotic movement—and eventually a panicked stampede that left 21 dead and 500 injured. The exact causes of the tragedy may never be certain, but scientists have developed a nimble computer model that can digitally reproduce such catastrophes—and perhaps help prevent them.

Crowd simulations help architects design safe public spaces. Current models mostly ignore human behavior, such as chatting or shoe-tying, because they introduce too much complexity and freeze up simulations. As a result, virtual people are treated as simple particles that repel one another while zooming toward their destinations. Such dehumanized models are useful, but they work only within limited parameters. A model that predicts how long it will take a group of pedestrians to cross a busy intersection, for example, generally fails to reconstruct the chaos seen when people try to escape a crowded room.

The new model is more accurate than any before it, and it does a lot more with a lot less. Created by behavioral scientist Mehdi Moussaid of the University of Toulouse in France and colleagues, the model values simplified rules of human behavior over complex physics—a first in modeling crowd dynamics. Although it doesn’t ignore physical laws, such as person-to-person energy transfer that help explain Love Parade-like “crowd-quakes," it does rely on two simple rules the team programmed into virtual pedestrians: Keep a safe distance from others and move through gaps between people.

Walk this way. In addition to dangerous situations, the new model reproduces phenomena seen in more typical scenarios, such as lanes of people that spontaneously emerge in sidewalk traffic.
Credit: Mehdi Moussaid/Dirk Helbing/Guy Theraulaz

Moussaid got the idea while chatting with a colleague about how players catch baseballs dropping from the sky—not an easy feat for first-timers. Researchers have discovered that people's brains don’t crunch complex equations to do it. Instead, our minds develop shortcuts based on experience. “We try to keep the angle of the falling ball constant in time. If we do that, the ball falls into our hands,” says complexity scientist and study co-author Dirk Helbing of the Swiss Federal Institute of Technology in Zurich. Such simple rules, called heuristics, can be transformed into simple mathematical formulas fit for computer models.

Get out! The new model faithfully reproduces the dynamics of crowds seen in real experiments, such as an emergency escape from a bottlenecked room.
Credit: Mehdi Moussaid/Dirk Helbing/Guy Theraulaz

After studying the patterns of pedestrians in videos, the team zeroed in on the rules the people used to get around and then converted them into heuristic formulas. When Moussaid and colleagues weaved the algorithms into an existing model and ran simulations, they discovered that the approach worked for a variety of situations. In simulations of simple hallway interactions, for example, lanes of pedestrians formed, as they do in real hallways (see video), the team reports online this week in the Proceedings of the National Academy of Sciences. Scenarios of emergency escapes from bottlenecked rooms also reproduced lifelike data (see video). And crowds similar to the Love Parade disaster produced dangerous waves of energy.

There’s never been a single crowd-dynamics model that reproduces such a large range of observed behaviors, says behavioral biologist Iain Couzin of Princeton University, who wasn’t involved in the study. “The tendency is to create one model per scenario,” he says. “This work is an extremely important step in pulling together our fragmented understanding. We’re now approaching a sort of unified understanding of human behavior in crowds.”

“Showing how simple local rules and mechanisms on an individual level can lead to patterns on a crowd level greatly helps disentangle cause and effect,” says complexity scientist Anders Johansson of University College London, who has previously collaborated with the team but wasn’t involved with the work. “It will ultimately help plan and manage crowded events in a better, safer way.” Helbing even thinks the model could help autonomous robots of the future efficiently navigate to their destinations, perhaps to deliver supplies within a busy hospital.