Many knowledge-based systems suffer from structural problems such as inefficiency rooted in overinformedness and inability to cope with the unexpected or exceptional nature of real-world data. Behaviour-based architectures are better suited for such problems, but are not yet widely applied, possibly because design strategies are not yet well established. In robotics, subsumption architecture has proven an effective framework for developing such systems. In this paper we suggest the techniques of subsumption architecture can be transferred to other areas of Artificial Intelligence, and present a project implemented in this fashion. The development strategies used and the types of problems approachable by this method are also discussed.
|Publication status||Published - 1992|