The Adaptive Route Advisor requires heavy memory resources to store the digital map and significant processing resources to compute an optimal route. However, we assume computational resources in vehicles will be limited in the near future. The client/server architecture shown in Figure 1 resolves this difficulty by offloading resource-intensive processes onto a remote server. This architecture also lets the routing system use information about the current traffic conditions, which would be available as a centralized service, as in the Intelligent Traffic Guidance System in Tokyo [3].
In the figure, portions in the current implementation are drawn in solid lines and planned extensions in dashed lines. The interface client is a resource-light process suitable for a vehicle's limited computational power that connects to the servers via a wireless TCP/IP connection. The route server receives requests from the client and uses the digital map to compute an optimal route according to preferences in the user model. Route requests may include short-term changes in the preference model to reflect unusual situations or corrections in the model.
The system initializes the agent with a default user model and refines this model with feedback from interaction with the interface. Future versions will also allow feedback from direct sensing of the driver's preferred routes using the Global Positioning System. Section 4 discusses these extensions in more detail.