An evidential distance for intelligent control

This work deals with an evidential approach to intelligent control. Control decisions are determined by analogy to the most "similar"' rules (prototypical decisions) in the knowledge base. These rules map control situations (represented by belief functions) into appropriate control actions. The main focus of this paper is on the derivation of a measure of dissimilarity, or distance, between belief functions corresponding to control situations. The proposed methodology is illustrated by simple examples from an application to the grip control of a multifingered robot hand.