Modeling Urban Panic

Schneier is reporting that Arizona State University's Paul Torrens has been developing a computer simulation to model urban panic. "The goal of this project is to develop a reusable and behaviorally founded computer model of pedestrian movement and crowd behavior amid dense urban environments, to serve as a test-bed for experimentation." The simulation tests behaviors from how a crowd flees from a burning car to how a pathogen might be transmitted through a mobile pedestrian over time among others.

Comments

You need a simulation for this?

[Anonymous Coward]

The simulation tests behaviors from how a crowd flees from a burning car

Hmmm... my guess is AWAY from the burning car.

Re:You need a simulation for this?

[evanbd]

Of course you do.

Any time you want to answer the question "What will happen to X in the event of Y?" you either need to try it, or you need a model. Your model ("crowd moves away") probably gives correct results, but not detailed ones. If you want more detailed results, with answers to more detailed questions (How fast? Which direction? How does it change with pedestrian density? How do obstacles matter?) then you need a better model.

If you want to improve pedestrian traffic, police response, crowd control... This model could be quite helpful.

Re:You need a simulation for this?

[phobos13013]

It could be, but I believe some parameters of the modeling AI need to be changed. Take a look at the massive crowd evacuating through the small aperture. The crowds on the fringe just stand there and pile up while the late-comers in the middle flow right through since the fringes dont move. In a real crowd (even under regular evacuation measures and not one in a "panic") would never go for this! People would constantly be trying to overcome the person not moving in front of them! The fringe individuals would immediately relocate to the center which is dwindling in length.

I don't think this model takes a lot of human behavior into consideration. One requirement could be that the individuals never stop moving, they will always take the available (open) path towards the exit even if it is not the straight line path which they are lined up in queue for. It also doesn't take into the consideration of a possibility of a trample situation where perhaps a threshold value of energy from a surging crowd overcomes the resistance of the small group of slower moving individuals in front. But of course, this is a great start for a complex computational issue!

Re:You need a simulation for this?

[trolltalk.com]

Crowds don't flee from burning cars any more than they flee from the scene of a burning building

They certainly do, if they are inside.

... and how the fsck do you get a crowd into a car? Round up a bunch of midgets at the local circus, stuff them in, toss some gasoline on them, and light it? That'll attract a crowd, for sure ...

Yes but can it model...

[billius]

a zombie attack? And if so, can it compensate for the differences between slow-moving George Romero zombies and fast-moving British zombies?

Re:Yes but can it model...

[morgan_greywolf]

Apparently, this can. (Linked in TFA)

didn't they do this...

[notgm]

didn't they do model this already in the Grand Theft Auto series?

step a. pedestrian looks at event.
step b. pedestrian throws hands in air.
step c. pedestrian runs away.
step d. pedestrian gets winded, approximately 1/2 block from event.
step e. pedestrian forgets event.
step f. pedestrian walks around aimlessly.
step g. (sometimes) pedestrian's head explodes, becomes event triggering new step a.

seemed pretty darn realistic to me.

I can answer one of these

[Malevolent Tester]

5) identify, if possible, the tell-tale signs of a peaceful crowd about to metamorphosize into a hellish mob;

Riot police. I've seen several demonstrations turn violent, and every single time it was preceded by riot police either attacking people (I've seen Metropolitan Police TSG hit a pregnant woman for talking back to them), herding people into an enclosed space and beating those who try and get out or baton charging a peaceful crowd.*

*This is not to say the police cause all riots, but they're certainly a factor in at least some of them.

Links to the Rendered Videos

[phobos13013]

Much more interesting than the Schneir description of the actual site in question. Here they have fully rendered videos from multiple vantages of the studies amongst other research topics of the professor.

Been there done that

[JustNiz]

My CS AI prof at University of Reading had done this about 15 years ago, following the tube train fire at Kings Cross, London.
He created a model of the station and passengers, programmed only about 6 simple rules into the movenent of each passenger, and found that the model pretty accurately recreated where they found the actual bodies in the station.

Number 4 is funnier

[spun]

4) design a mall which can compel customers to shop to the point of bankruptcy, to walk obliviously for miles and miles and miles, endlessly to the point of physical exhaustion and even death; Now that's what I call science! Now, if only we could design a website that compelled people to browse, even to the point of losing their jobs, reading obliviously to their need to shower, posting endlessly to the point of mental confusion and even complete retardation... Oh, wait...

There's more than movement, too

[blueZ3]

Those people on the outside are pressing against the folks in the middle and people at the back are pressing forward. The nice stream shown doesn't appear to account for this, especially as none of the sims are crushed, trampled, or otherwise flattened in the mad rush to the door.

IRL, people on the outsides frantically push their way toward the exit, creating pressure on those in the center that frequently results in a crush of bodies that this model doesn't seem to model very well. If you've ever been in a situation where the crowd pressure to pass through a bottleneck is so strong that you can't move backward, hold still, or even effectively resist the rush, you know what I'm talking about.

This model seems to be a "in a perfect world, where the panicked crowd moves cooperatively and generally in an orderly fashion towards the exits" kind of model. It's hard to see how that's very useful in the context suggested (panic response).

Direct link

[Dan100]

to the research. Even better, the ASU press release went out seven months ago.

What about a dynamic environments...

[moncho]

This completely reminded me of the opening of the London Millennium Bridge in London (crossing the Thames) were the bridge (nicknamed the Wobbly bridge) began to sway due to a few pedestrians who, by happenstance, inadvertently stepped in the same direction at the same time, causing a slight sway which on the rebound caused a few more people to step into the same direction, causing further swaying, increasing the effect w/ every oscillation. This effect is known as Synchronous Lateral Excitation. The funny thing is that each step, even several in synchrony, have negligible effects on bridge stress models... it was that this particular sway happened in such a way that forced more pedestrians on the bridge to step INTO the direction of the sway, continuing until most everyone on the bridge (up to 2,000 pedestrians) were contributing energy to the sway. The aforementioned is an instance of an unexpected design flaw due to inadequate modeling, and one can always come up w/ such instances, but these are meant to be learned and avoided... not repeated.