Neural Network Simulation as a Computer- Aided design Tool for Predicting Wayfinding Performance

Complex public facilities such as libraries, hospitals, and governmental buildings often present problems to users who must find their way through them. Research shows that difficulty in wayfinding has costs in terms of time, money, public safety, and stress that results from being lost. While a wide range of architectural research supports the notion that ease of wayfinding should be a criterion for good design, architects have no method for evaluating how well their building designs will support the wayfinding task. People store and retrieve information about the layout of the built environment in a knowledge representation known as the cognitive map. People depend on the information stored in the cognitive map to find their way through buildings. Although there are numerous simulations of the cognitive map, the mechanisms of these models are not constrained by what is known about the neurophysiology of the brain. Rather, these models incorporate search mechanisms that act on semantically encoded information about the environment. In this paper the author describes the evaluation and application of an artificial neural network simulation of the cognitive map as a means of predicting wayfinding behaviour in buildings. This simulation is called NAPS-PC (Network Activity Processing Simulator--PC version). This physiologically plausible model represents knowledge about the layout of the environment through a network of inter-connected processing elements. The performance of NAPS-PC was evaluated against actual human wayfinding performance. The study found that the simulation generated behaviour that matched the performance of human participants. After the validation, NAPS-PC was modified so that it could read environmental information directly from AutoCAD (a popular micro-computer-based CAD software package) drawing files, and perform'wayfinding'tasks based on that environmental information. This prototype tool, called AutoNet, is conceptualized as a means of allowing designers to predict the wayfinding performance of users in a building before it is actually buil