Discussions around the office about agent-based modelling are becoming interesting. One question arising from the discussion is whether Pacman an agent-based model?While packman himself is controlled by the user, the ghosts are programmed as individuals with the task to find and destroy him.
There are a number of agent-based modelling toolkits available for the creation of 3D agent-based models (e.g. NetLogo, Repast).However there is difficulty in incorporating geographical information into such models. On solution is potentially importing such geographical information into a 3D modelling package and creating agents directly within such a system.
While the first movie highlights the first steps in developing 'Ant Like' behaviours on a surface, their second movie displays a degree of intelligence in their agents.The agents are now aware of the environment around them and each other, as such they avoid collisions while wandering around the surface:
While this is work in progress, Andy is developing this work further hoping to develop a 3D city model with agents such as cars and pedestrians interacting with their environment and with each other.
For more information on the creation of 3D models and updates of the work its well worth checking out the Digital Urban blog.Andy is planning on writing a tutorial on the system as soon as it is more advanced.
On a side note one can also loosely couple an agent-based model with 3D Studio Max, for example T. Narahara of HarvardUniversity, Cambridge, MA it explores the integration of NetLogo with 3D Studio Max. Under a paper entitled ‘Enactment Software: Spatial Designs using Agents-Based Models.’
The Agent-based Environment in NetLogo (left), and the 3-D Visualization (right)
Over the last few weeks Duncan Smith and I have been working on paper entitled “Fine Scale Modelling of the London Housing Market” for the GISRUK 2008 conference.
The motivation behind this research comes from the lack of fine scale property data within England. Unlike other countries, England has no national cadastre on housing attributes (such as size, type and age) which restricts application of GIS in planning. While new datasets are becoming available, we have become interested in exploring if combining recent datasets, such as OS address data and Land Registry transaction data can begin to fill this data gap? And if this is the case, could a fine scale housing database developed from these datasets be used to improve house price modelling?
Isle of Dogs Housing Density 3D Map.
While at present this is work in progress, the presentation outlines our current work and initial findings while the paper suggests future work, such as the creation of an agent-based model exploring residential choice.
Crooks, A. T., Smith, D. A., and Theseira, M. (2008), The Fine
Scale Spatial Dynamics of the Greater London Housing Market, in
Lambrick, D. (ed.), Proceedings of the 16th Geographical Information Systems Research UK Conference, UNIGIS, Manchester Metropolitan University, England, pp. 117-124. (pdf)
Mike Batty at CASA has been working on a Land use transport model for London. The model simulates the location of the residential population as a function of this employment, floorspace and generalised travel cost. The model is currently, a partially constrained spatial interaction/residential location model disaggregated by four modes of transport – road, heavy rail, light rail (tube and DLR) and bus, with walk-cycle-other the fifth residual mode.
The model is highly visual making the entire modelling process as transparent as possible. Moreover to verify the data and to interpret the spatial structure of the metropolitan area is made easier through the visualisation of maps and graphs.
If you are interested, Mike has recently created a movie (click here to view the movie) that lasts for about 12 minutes and takes you through the sequence of stages from input data to calibrating the model to using it for predictions.
Interface to the model
While there is no definitive paper about the model at present, the movie (click here to view) and the following working papers provide an good insight into the model.
Often when I talk about agents and Agent-based models I give the example of cars on a simple road, with each car being an agent.Each car follows a simple set of rules: it slows down (decelerates) if it sees a car close ahead, and speeds up (accelerates) if it doesn't see a car ahead.
The model is simple but demonstrates how traffic jams can form (emerge) without any accidents, broken bridges, or overturned trucks. Furthermore it show how no "centralized cause" is needed for a traffic jam to form.
The mathematical theory behind these so-called "shockwave" jams was developed more than 15 years ago using models that show jams appear from nowhere on roads carrying their maximum capacity of free-flowing traffic – typically triggered by a single driver slowing down.
The theory has frequently been modelled in computer simulations, and seems to fit with observations of real traffic, but has never been recreated experimentally until now.Now a team of Japanese researchers has recreated the phenomenon on a test-track for the first time by putting 22 vehicles on a 230-metre single-lane circuit (see the movie below).
Drivers were asked to cruise steadily at 30 kilometres per hour, and at first the traffic moved freely. But small fluctuations soon appeared in distances between cars, breaking down the free flow, until finally a cluster of several vehicles was forced to stop completely for a moment.That cluster spread backwards through the traffic like a shockwave. Every time a vehicle at the front of the cluster was able to escape at up to 40 km/h, another vehicle joined the back of the jam.
The full article can be read in New Scientist (click here).
Discussions around the office about agent-based modelling are becoming interesting. One question arising from the discussion is whether Pacman an agent-based model? While packman himself is controlled by the user, the ghosts are programmed as individuals with the task to find and destroy him. 
