Agent-based modelling of pedestrian movement

Abstract

Agent-based modelling has a long history of application in human movement simulation, but little has concentrated on the validation of the model simulations against measured navigational patterns.  There has been much research on modelling group behaviour, especially in the context of designing fire escapes and crowd control.  These models are ideal when the behaviour is in large aggregations, but have little in common with less densely populated areas.  These short-falls are almost certainly due to the lack of visual control in most agents.  By contrast, space syntax has shown good correlation between graphs based on line-of-sight and actual human movement, by concentrating entirely on the visual relationships within spaces to account for movement.  Thus, it seems imperative that agent simulations of human behaviour should include a sight-based model.  My research aims to bring together the disparate knowledge domains of agent-based simulation and space syntax.

Figure 1 : Visual agents in a gallery space (screenshot from the EVAS program)