A behavior-based control architecture

The presented control architecture is designed to support the development of intelligent control systems that are connected to the real world via sensors and actuators, and deal with dynamic and partly unpredictable environments, for instance to control an autonomous mobile robot. With quite an implementational effort, existing control architectures allow us to implement navigation facilities on a certain mobile robot. However, additional criteria that interfere with the navigational movements can usually not be added subsequently. Note that this situation occurs, e.g. if active sensing is requested, a handling device is added or other additional requirements arise. The purpose of the presented architecture is, first to allow easy implementation and second to obtain a modular control system that displays a kind of superposition characteristic. Considering a typical behavior based architecture, we focus on the method to coordinate the different mechanisms. The question is, how to combine commands or preferences of several running mechanisms. In our approach, we solve this merging problem by using a utility based approach. Each mechanism continuously evaluates sensory data with respect to its special sub-task in order to rate all actions of a certain actuator. These ratings result in a utility function, that also includes the priority of the performed sub-task of the mechanism. A fusion component then merges these distributions for each actuator, following a theorem of R.L. Keeny and H. Raiffa (1976).