bart van greevenbroek. authors the paper particle swarm optimization algorithm used with pso ...

23
Controlling the movement of crowds in computer graphics by using the mechanism of particle swarm optimization Bart van Greevenbroek

Post on 19-Dec-2015

214 views

Category:

Documents


0 download

TRANSCRIPT

Controlling the movement of crowds in computer graphics by usingthe mechanism of particle swarm optimization

Bart van Greevenbroek

Overview

Authors The Paper Particle Swarm Optimization Algorithm used with PSO Experiment Assessment conclusion

Authors

Ying-ping Chen Ying-yin Lin

The paper

Published in 2009

Particle Swarm Optimisation

Developed by Kennedy and Eberhart Published in 1995 Inspired by flocking of birds and

schools of fish Solution is modeled as a flying

particle in a hyper-plane

Particle Swarm Optimisation (2)

= velocity of particle i at the next timestep= the weight for the previous velocity= the best position where this particle had been= the overall global best position ever achieved by the swarm= cognitive and social parameters, deciding the influence of Pbls and Pbgs= random factor, to produce varied paths.

= position of particle i at the current timestep.

Particle Swarm Optimisation (3)

Particle Swarm Optimisation (4)

Every particle has an objective function, which can influence a and .

It does not take obstacles into account, making PSO incompatible for crowd simulation in its current form.

Algorithm with PSO

Each pedestrian is considered a particle in 2d space, with position pi = [pix , piz] T a direction Di = [Dix ,Diz ]T and a speed S.

Direction function

and are unit vectors.

Position function

The new position is determined by the direction and the speed.

Method

Speed is updated to the inverse of the objective function. This varies the pace of each person.

If a particle approaches an obstacle, the speed will be slower due to greater objective values.

Exponential cost function

Exponential Obstacle Model

Objective Cost Function

= balancing factor that decides the balance between avoiding obstacles and reaching the goal.

= low if the cost to the goal is high.

= the object that has the highest cost (closest obstacle)

Probability

is a constant factor that can influence the probability of the new position being accepted.

Probability (2)

Experiments

A number of experiments were performed

To show how bad this method is.

Experiments

No Details on the implementation are given

No system specs No performance No way to compare with other

methods Except the movies which show very

non-human like behavior

Engrish

What can we learn from this?

Swarm Intelligence is NOT a good way to model human behavior

Other predictive methods look much nicer.

The desire to make something general will not work when you have specific situations requiring specific solutions.

Occilations

In the abstract the authors state that they want to avoid oscillations which works with the original PSO. But the examples shown oscillate like ants

What were they thinking?