Controlling Individual Agents inHigh Density Crowd SimulationN. Pelechano, J.M. Allbeck and N.I. Badler (2007)

OutlineIntroductionRelated WorkThe ModelResultsConclusionsAssesments

The AuthorsN. Pelechano

◦ Assoc. Prof. at Catalunya University.◦ Crowd simulation, real-time 3D, human-

avatar interactionsJ.M. Allbeck

◦ Assist. Prof. at George Mason University.◦ Animation, AI, physcology in crowds

N.I. Badler◦ Professor at University of Pennsylvania◦ Computational connections between

language and action

IntroductionA model for High Density

Autonomous Crowds (HiDAC)◦Natural, realistic crowd simulation◦Handle high density◦Adapt to dynamic changes

IntroductionHybrid approachPhysical forces with rules:

◦Physiological (locomotion)◦Psychological (personality, panic..)◦Geometrical (distance, angles..)

Two levels:◦High level – global◦Low level – local

Related WorkHelbing’s Social Forces model

◦Particle simulations , Oscillations◦Extensions exist – real-time problems

Rule-based models, i.e. Reynold’s◦ Realistic, for low-medium density◦Avoid individual contacts

Related WorkCellular Automota models

◦No contact between agents

Higher level navigation◦Roadmaps◦Potential Fields◦Cell and portal graphs

Related Work

The Model - Overview

High Level Module

Modeling Crowd and Trained Leader Behavior during Building Evacuation, by Pelechano and Badler. (2006)

Low Level ModulePrevent oscillationsCreate bi-directional flowsQueueingPushingAgents falling and act as

obstaclesPropogate panicExhibit impatienceReact to dynamic changes

Low Level ModuleMovement of an agent

Low Level ModuleThen, position is:

◦α : agent will move or be pushed◦v : velocity ( <= Vmax), constant a◦β : priority value to avoid fallen agents◦r : result of repulsive forces ◦T : time step

Forces: Avoidance

Forces:Avoidance

• D : viewing rectangle• Increase/decrease based on density

• Weights: • d: distance between agents• o: orientation of velocity vector

Forces: AvoidanceBi-directional flows with right

preference and altering rectangle of influence

Forces: Repulsion

•λ : Priority value between agents and walls/obstacles• Walls > Agents

Shaking ProblemStop moving if:

◦Agent is not in panic◦Repulsion against the agent

Can still be pushed forward.

Waiting BehaviourAllows queueingDisk of influence

◦Depends on desired behaviour

Pushing BehaviourPersonal space (disk)

◦I.e. Low for impatient agentApply collision response force

Falling AgentsWhen pushing forces are high

Becomes an obstacle

No repulsive force

Panic PropagationHigh-level module

◦Communication between agents

Low-level module◦Agent detects density or pushing

Dynamic changes and bottlenecksHigh-level module

◦Supply alternative paths

Results85 room environment

Simulation only:◦25 fps◦1800 characters

Simulation and 3D rendering◦25 fps◦600 simple 3d human figures

ConclusionsAbility to simulate low-high

density◦Panic and calm situations

New and natural behaviours◦Pushing, queueing, falling agents...

User needs to define parameters for different environments/situations

Assesments – The paperLocal methods/behaviours

◦Clear explanation◦Supported with figures and results

Experiments & Results◦Rather scattered◦One or few comparative tests◦Could be more

Assesments – The methodNo problems with the model?

Behaviours and the model depend also on high-level module◦Limited adaptability◦Gaps in the method explanation

Assesments – The methodPerformance

◦25 fps, 600 human figures◦Enough for simulations and/or

games?

Applicability◦Rather limited ◦Would serve for industrial

applications

Assesments – The methodIncorporate global and local

approachNatural in high density

◦Individual contacts/interactions

Globay wayfinding◦Shortest path◦Maybe deliver another approach

Roadmaps, corridor maps

Assesments – The methodLacks prediction/anticipation

◦ A Predictive Collision Avoidance Model for Pedestrian Simulation, Karamouzas et al.(2009)

Able to handle high density◦ Morphable Crowds, Eunjung Ju et al. (2010)

Integration of a personality model◦ How the Ocean Personality Model Affects the

Perception of Crowds, F. Durupinar et al. ( 2011)