Pedestrians Walking - Crowd Simulation

[dumbzebra]

I guess it's something like:

a)Pedestrians walk straight forward

b)if they get into the red area, they change course slightly

c)if they get into the yellow area, they change course drastically

[pzmtuthcvpvl]

@dumbzebra seems the advancing from the left had a yellow area as you describe but the others did not care

[birlibis2juggler]

Wall of death fail!

[justinlovespastry]

Why would two giant groups of people clash together at the same speed? XD

[TheallCreator]

@justinlovespastry Pedestrian crossings

[spectrex1]

would it be too much to ask to give details of the mathematics behind this model? This is amazing i'm thinking about trying one of my own in matlab, but a starting point would be nice :)

[badmephisto]

@spectrex1 Sure: Define P1 to be center of agent. Define P2 to be slightly ahead of him. Danger(p)= e^(-a*(|P1-p|+|P2-p|)). Where p=(x,y) is a 2d position and a is some scalar constant. Take derivative w.r.t. to p to get the gradient of this field at every point. An agent acts to minimize the danger, so in addition to his regular velocity, add a term that points opposite to the gradient of the total field caused by all agents.

[spectrex1]

@badmephisto thanks much appreciated, seems alright, i'll let you know how i get on and what marvellous things i create :)

[badmephisto]

@spectrex1 You can find code at pastebin.com SLASH meQcwfWw

The method of interest is gradDangerAt(x,y)

[TheRedComa]

Watch this video, and in you mind, say, "Excuse me, pardon me, excuse me, pardon, pardon, coming through."

[dorsenstein]

So this is how they programmed the matrix XD

Excellent job. I love all of your programming videos and I hope you will make some more.

[gresach]

Very pretty, but in real life when two people are trying to pass one another, they very often jiggle from left to right to figure out which side to pass. I didn't see that happen at all.

[badmephisto]

no that definitely DOES happen sometimes in the simulation! It was one of the reasons I liked it so much! Indeed I don't see it happen exactly in this demo... But one significant disadvantage I think to this is that they can't side step. People side step ALL the time...i.e. change COM position without affecting the direction. These guys have to change the direction first and then travel along it :(

[weetbixiron]

Wow I remember seeing this a long time ago, and just recently I wanted to watch it again. Found it in related while watching your NN video :)

[donmutti0]

I'd like to see potential field animation at last example

[RedSword69]

Impressive

[decay]

very nice what language is this written in?

[badmephisto]

Python, ftw.

[KevinLounsberry]

cool

[patrick1020000]

very cool

[josephohsnap]

this was great.i am a little sad that the more i look, the more ideas that i have in my head i find out are already done. :( i am not as original as i thought.

[tidusvssora]

as long as you have the idea and they aren't ased off of anything that you've seen, then it is original. it doesn't matter if it's already been done, the idea is yours and thats it. nothing else matters.

[serloinz]

@josephohsnap tell ma about it man.. bloody tell me about it :)

[hyssp]

Wooo!! Great!! have you seen a little app called "bugtro"? Im pretty sure you will love it :) Congrats!!!

[commentitout]

Very nice video! Just a question, its written in pygame? How CPU intensive is the calculations?

[badmephisto]

yes its pygame. And its pretty efficient unless you are drawing the potential fields... (which is an n^2 operation in number of pixels). Otherwise i can have at least like 50 or 100 people walking without too much lag.

[KnowledgeReference]

You have loads of the same intrests as me :D

Rubik's cubes, programming, pen spinning.

[Loktantra]

HI..v nice simulation. Can you elaborate the mathematical function and how does it work? Thank you

[eFFeeMMe]

That was very interesting. Thanks for taking the time to explain!

[eFFeeMMe]

Good job! Can you post more information about how this works?

[badmephisto]

well every agent has an associated "danger function" that has elliptic shape extending in the direction he faces. Every agent in addition to walking to his target direction is trying to minimize the total danger at his position. So you take the sum of all gradients of all danger fields at agent's point, and this will give you direction of greatest danger increase. So just move opposite to that + add the vector of motion towards target

[DonZabu]

That's darwinbots.

[badmephisto]

No It's not. :) My program has nothing to do with evolution, DNA, or natural selection. But yes, my agents are also circles with a direction line :)