Rod's Agent Based Pedestrian Model
From GeoMod
Agent Based Pedestrian Model:
Purpose:
The purpose of this writeup is to discuss the development of an agent based pedestrian model.
Discussion:
Introduction: Agent based pedestrian models operate on the finest geographic scale. The scale of pedestrian agent models is often very large. The movements of agents ranging in scale from the inside of buildings to that of multiple city blocks have been explored. Many agent based pedestrian models simulate such places as stadiums, train stations any place where the motion of pedestrians may affect, safety or sales or other conditions(Hoogendoorn, et al.,2002). ABM pedestrian models are composed of a series of finite state machines and a simulated environment. A set of prescribed behaviors are the machine output. Current environmental conditions are the machine input. Transition rules define the matching of machine output to environmental input(O’Sullivan, et al., 2000 ).The simulated environment is an abstraction of the real world. The geometry of the environment is represented in the model as areas accessible or not to the pedestrian agents. These may include streets, building, walls, or other barriers. Pedestrian agents, like their human counterparts posses the desire to reach their travel destinations. While expressing this desire by moving through the simulation agents react to the geometry of the environment and interact with other agents. Patterns of behavior emerge from the agent-environment, and agent-agent interactions during the simulation. The use of agents provides for the study of emerging global patterns flocking and swarming as a result of many individual actions( Therakomen, 2001).
Theory:
The motion of pedestrian agents is based upon the work A Mathematical Model for the Behavior of Pedestrian Helbing (1998) refers to his generic model of pedestrian movement as a social force model in which each of these features is associated with a force that pushes the walker in a particular direction. In general we might think of movement to a new location as being formed from
[new position] = [old position] + [desired position] + [geometric repulsion] + [social repulsion] + [social attraction] + ε
(Batty,2003)
Conclusions:
Setup is the first concern I would like to address. The development of the model is in its early phases. It is my first attempt at agent based pedestrian modeling. It is also the most complex piece of
software I have written. The idealizes space was designed to demonstrate the navigation of pedestrian agents around obstacles and to preview the formation of emergent patterns. It does not exist and
probably will not in such an idealized and simple manner. More realism is needed in the simulation space. Let us address the terms of the social forces model along with the pros and cons of my
implementation as well as what I would like to happen in the future.[old position] My model has no memory of old position. The current conditions prescribe the agents next movement. Humans have memory and so should agents. If location preferences were developed among the pedestrians, a form of social attraction term, some method should be developed for the storage of locations preferences. [new position]The implementation in this model has much to be desired. The agents have a hard enough time negotiating a few regular polygons. The implementations of Delaunay triangle edges between points would simplify movement for each agent. These edges chained together will provide the shortest free path to a desired position. A module is available in the SciPy package I should give it a try.[geometric repulsion] Appears to be the main focus of this model. I am satisfied with the model action so far. There is room for improvement. The rate at which pedestrians should change with velocity and proximity to a geometric barrier needs to be variable. The current implementation is either on, avoid wall or off . This could be implemented by a series of operations similar to the zon_col method. [social repulsion],[social attraction] again on or off agents need to be aware of the formations of groups and choose to join or disperse. Agents also need to express conditional behavior when approaching another pedestrian. Ideally they should veer to avoid. Working this out will probably take a little while I have seen example of this behavior outlined in the literature but no code examples. in the end should be tuned after investigation as to the real meaning of the term, but so far Iam satisfied. In conclusion agent based pedestrian modeling is a very complex subject. It combines Physics, statistics, psychology, and GIS all into one package. My implementation falls very far from the current tree. But I realize you have to start somewhere.
Operation:
Lets begin the exploration of my model
It has been divided into operational sections
- mainagent This is the main operational class of the model
- gamespace This class is responsible for setting the simulation space
- targetgen This creates target destinations for the pedestrians
- pedagent Pedestrians are created and movement transition rules are here
- recorder This class writes the locations of pedestrian agents out to a data file
References:
[Batty,2003]
Michael Batty:Agent-Based Pedestrian Modeling, January 2003, pp.1-28
Centre for Advanced Spatial Analysis University College London http://www.casa.ucl.ac.uk/working_papers/paper61.pdf
[Batty,et al.,2002]
Michael Batty, Jake DeSyllas, & Elspeth Duxbury The Discrete Dynamics of Small
Scale Spatial Events: Agent-Based Models of Mobility in Carnivals and Street Parades
10 July 2002, pp.1-45 Centre for Advanced Spatial Analysis
University College London http//www.casa.ucl.ac.uk/working_papers/paper56.pdf
[Helbing,1998]
Dirk Helbing: A Mathematical Model for the Behavior of Pedestrians
15.arXiv:cond-mat/9805202 v1 18 May 1998, pp.1-21
II. Institut f¨ur Theoretische Physik
Universit¨at Stuttgart http://www.theo2.physik.uni-stuttgart.de/helbing.html
[Hoogendoorn,et al, 2002]
Serge P. Hoogendoorn, Winnie Daamen, Piet H.L. Bovy: Extracting Microscopic Pedestrian Characteristics
from Video Data Results from experimental research into pedestrian walking behavior, TRB 2003 Annual Meeting CD-ROM,
pp. 1-15, Transportation and Traffic Engineering Section Faculty of Civil Engineering and Geosciences Delft University of Technology
[Schelhorn,et al.,1999]
Thorsten Schelhorn, David O' Sullivan, Mordechay Hacklay, Mark Thurstain-Goodwin
Streets: An Agent-Based Pedestrian Model, April 1999, pp.1-14
Centre for Advanced Spatial Analysis University College London
http://www.casa.ucl.ac.uk/streets.pdf
[O’Sullivan ,et al., 2000]
David O’Sullivan, Paul M. Torrens: Cellular Models of Urban Systems,
June 2000, pp. 1-13, Centre for Advanced Spatial Analysis University College London, http://www.casa.ucl.ac.uk/cellularmodels.pdf
