DIMACS logo

Agent Based Modeling of Crowd Behavior

Josh Zukoff
Mentor: Cliff Behrens



General Information Presentations Weekly Updates


The focus of this research is agent based modeling (ABM) of crowd behavior. We will use NetLogo to model crowding and the behavior that emerges from the interaction of individual human agents in crowded conditions. The emphasis will be on representing in models realistic assumptions about human cognition and emotion in situations such as emergency evacuations from built spaces, and comparison of the relative quality of knowledge obtained through crowd-sourcing and expert opinion sampling.


General Information

Name: Josh Zukoff
Mentor: Dr. Cliff Behrens
Topic: Agent Based Modeling of Crowd Behavior in Emergency Situations
Office: CoRE 434 Rutgers University: Busch Campus
Email: josh dot zukoff at gmail dot com
School: Colgate University

Presentations

Introductory Presentation
Final Presentation
Sample Netlogo Applet (Wolf Sheep Predation)


Weekly Updates

Week 00

Read through literature provided by Dr. Behrens to get acquainted with the topic. Also worked on thinking of a specific direction for my research based on what I had read.

Week 01

Continued reading papers and worked on both the powerpoint presentation and a basic website design. Began work on a basic netlogo model with a simplified map and agents without much intelligence.

Week 02

Worked on implementation of my basic netlogo model. Have achieved agents which can successfully exit a room and then a building albeit without much intelligence. Goal for the week is to implement some sort of basic group interactions. First meeting with Dr. Behrens on wednesday: Discussed importance of having a more realistic map to work on and some more ideas of path that should be taken with the project. I have finally been able to import a map of a stadium which I made and have had a bit of success with the movement of my agents. Nothing complicated has been implemented as of yet.

I have attached an applet so that my progress can be tracked.
Week 02 Netlogo Simulation

Week 03

Began to complicate my model a bit. Have two different options for maps, one with two exits and another with 4. Also have added spots where people from upper levels would enter into the concourse. Have added in a "global notification" system which tells everyone where the closest exit is. I have implemented collision avoidance amongst my agents. Met with Dr. Behrens again, immediate goals are to implement some sort of social comparison function, reevaluate and document all assumptions I have made, remove any "forcing" of the model and rely only on most basic actions. I took out my low level decision making processes for the escape of the building and opted instead for a modified flocking algorithm. This algorithm is meant to simulate the flocking of birds but works quite well to simulate naive crowd behavior. My next goal is to modify the flocking algorithm further to allow for the agents to intelligently move towards a destination. The next step will be to again modify the flocking algorithm to have more precision over the creating of flocks. Eventually I would like flocks to form when based on a number of factors, the agents would form a group.

Week 04

Began the week by reading a paper describing a model written by my mentor Cliff Behrens. This model was of American Soldier - Iraqi interactions. I am reading the paper to gain insight into the integration of a classification tree algorithm into a Netlogo model which may prove useful in my research. I have given my agents the ability to both search for the exit and learn the location of the exit from agents around itself. An interesting thing to note is that in my naive model, allowing the agents to search for the exits actually reduces exit time as compared to endowing all agents with knowledge of the nearest exit at the start of the evacuation. I have also encoded two different types of information sharing. In one scenario, most of the information is scared within a group and in the other information is shared locally regardless of group formation. I have implemented to role of a leader within a group. The leader is the agent that communicates across flocks and shares his or her knowledge with the rest of the flock.

Week 05

This week I will be honing in on a goal for my research, in addition to ironing out any kinks in my model. Additionally I will be attempting to complicate the model in a number of ways. I have implemented police officers which convey the location of the nearest exit. Have implemented a preference for a particular exit. This was done by giving each agent an exit number and then having them prefer to look for that exit rather than the nearest. Also began learning distributed version control and using it with my project.

Week 06

I have fixed some bugs in my exit preference code and have added the ability to turn this on and off. I still need to come up with a well defined goal for this project, that will be what I hope to accomplish this week. I have met with Dr. Behrens again to discuss where I want to wind up with my research. We have decided that it would be best to focus in on how various grouping scenarios affect evacuation time.

Week 07

This week I will be running a number of scenarios to try to find any interesting behavior resulting in my model. I will also prepare for my final presentation which will be given on thursday. Have fixed the grouping behavior in my model. Not quite perfect but its better than it was previously.

Week 08

This week I will be working on my write up report. Additionally I will begin work on a new project with Dr. Fiorioni and James Leslie.


Top