July 17, 2006

Getting crowded
Simulation helps troops learn how civilian throngs act

By Michael Peck
Special to the Times


If sticks and stones can break your bones, why provoke a mob?

In the hair-trigger world of stabilization and peacekeeping operations, troops frequently encounter locals with wicked tempers and a fondness for throwing things.

Incorporating mob behavior — from angry protesters looking for a fight to nosy bystanders who just want to see what the fuss is about — into defense simulations would seem to be a no-brainer.


“You don’t want to train in urban operations when there are no people around, or the people are ignoring you or are not responsive to what you are doing,” said Rick McKenzie, an associate professor of electrical and computer engineering at Old Dominion University in Norfolk, Va. “You are not going to go into Fallujah and not be noticed.”

But defense simulations have been unable to effectively model civilian crowds. So McKenzie devised Crowd Federate, a model that will add a crowd component to defense simulations.

“The intent is to provide a real-time, realistic, psychologically based crowd model to provide interactions with control forces,” he said.

Based on extensive psychological research, Crowd Federate works at several levels. At the smallest, the model tracks individual people as they travel around a city. The model’s psychological aspects kick in when 10 or more people congregate.

“There are different types of groups,” McKenzie said. “There is the protester group, which protests for a cause. They’re the ones holding the banners. The agitator group is there to cause trouble. The bystanders are just there and don’t want to get involved. Then, there is the curious group that will move toward anything interesting and stick their noses in. If something violent should erupt, they will probably run away.”

To some extent, the behavior of the military forces will determine the response of these groups. The very presence of troops may ratchet up tensions, as will shooting into the crowd. But other factors influence crowd behavior, and many can be adjusted by simulation operators. One key sets the overall crowd emotion level.

This is expressed in nine levels of aggression — “from being afraid all the way up to lethally violent,” McKenzie said.

Crowd Federate has two cultural variables: One sets the aggressiveness of the local population, the other sets the troops’ awareness of local customs. The most volatile combination is an aggressive culture and a culturally insensitive military force.

There are also environmental factors. Terrain affects the aggressiveness of crowds and their tendency to mass.

“If terrain is very difficult or steep, then it’s less likely to motivate them. If the terrain is flat and there is no problem for them to gather and mass in an area, they are more likely to rally more aggressively,” McKenzie said.

And while Crowd Federate doesn’t yet model weather effects, it may do so in the future.

“We know [crowds] are less motivated if it’s cold and rainy,” McKenzie said.

While the model will mostly be used in urban-operations exercises, it could be plugged into any exercise that involves crowds of civilians, such as a prisoner-of-war camp in an open field.

Only the behavior of civilians is modeled in Crowd Federate. Military forces will be controlled through whatever military simulation that is tied into it.

Research-based crowds

The model is based on extensive psychological research. McKenzie uses a team of psychologists that reviews existing literature, creates surveys and conducts interviews.

Interviewees include troops with experience facing crowds, as well as members of police forces from New Orleans and cities in California.

“From looking at videos of crowds and from interviews, we determined the frequency of behavior for a given crowd in a given context,” McKenzie said.

Researchers are also taking advantage of crowd-behavior workshops sponsored by the Defense Department’s Joint Non-Lethal Weapons Directorate.

Crowd behavior had not truly been reflected in military simulations before Crowd Federate, McKenzie said.

One popular model is Culture Sim, formerly known as Clutter, but it features only general population behavior, such as automobile traffic or mothers taking their children to school.

Other simulations tend to represent crowds as monolithic groups capable only of broad behavior, said John Camp, a computer scientist at the Air Force Research Laboratory and the government program manager for Crowd Federate.

“Either they’re aggressive or they’re not aggressive,” he said. “They’re just single entities with a central mass point. They don’t mill about.”

One difficulty in simulating crowd behavior is the less regimented structure of civilian life, McKenzie said.

Indeed, McKenzie said he found the literature on crowd behavior to be sparse.

Mogadishu lessons

Crowd Federate has been validated by creating reference scenarios of real events such as the 1993 ambush of U.S. forces in Mogadishu, Somalia. Researchers attempted to find out as much as they could about what happened, including what orders military forces had. The model was run with differing crowd behaviors until the historical outcome was recreated.

One area that McKenzie is working on is modeling the impact of nonlethal weapons, using injury models developed at the University of Pennsylvania. The module would figure out whether individuals were uninjured, lightly wounded or mortally wounded.

McKenzie estimated that Crowd Federate has cost $2.5 million over the past four years. The work is being done under the auspices of the Virginia Modeling and Simulation Center, a research center affiliated with Old Dominion.

The Crowd Federate project is sponsored by the Defense Modeling and Simulation Office, Air Force Research Laboratory and Joint Forces Command. The Air Force is managing the project but will turn it over to U.S. Joint Forces Command.

Ellie