Source: http://cumincad.architexturez.net/system/files/pdf/sigradi2004_310.content.pdf

Introduction

“object-oriented approach for simulating agent-based egress behavior”
web app
existing models of crowds deal with average speed or flow. this models “crowds as malleable, moving masses” to create “a more refined simulation”

individual behavior becomes irrational during panic situation
behavior of pedestrians in architecture – Okazaki 1979
pedestrians in urban planning – Jing 1999
GIS agent-based systems – Batty, Desyllas, and Duxbury 1999
agents in virtual environment – Penn and Turner 2002
boids offer realistic animations of group behaviors – Reynolds 1987

Object-orients programming approach to simulation

implemented in Java
GUI shows simulation stats, simulation time, and dynamic graphs

System Architecture

Divided into 4 phases: Initialization, Creation, Simulation, and Ending
Initialization phase – builds objects of Avatar Simulation class

pg 310

Scenario object contains all objects in scene: Person, Wall, Goal, Scaled Canvas, Avatar Box, and Vector objects.
Input file defines scenario

Creation Phase – begins when applet is launched. Creates Scenario object, Graph Window, Scaled Canvas
Person has Goal, Behavior, and Objective

Simulation Phase begins when user presses ‘run’
updates display every 10th of a second
Ending Phase user can see graph and simulation parameters

Implementation

collision avoidance
vary “parameters such as goal selection wait time and steps taken between decision points”
characteristics of ‘escape panic’ in people – Helbing
“e simulation also takes into account that interaction in a panic situation becomes physical, pushing behavior
is observed and fallen people act as obstacles.”
input file specifies min and max speed as well as ‘wait time’ “when an avatar becomes aggressive and starts pushing other avatars and trampling occurs. At the start, the avatar does not immediately move towards the shortest exit. Instead it moves towards a favorite goal that is pre-assigned in the input file. In real time scenarios people tend to move towards familiar exits. The simulation also takes into account false exits,
because in a real time scenario people tend to go towards a door that is not an exit. The avatars in the simulation move to another exit after reaching a false exit.”

line of sight and subgoals are used to avoid obstacles.

pg 311