Evolutionary Learning of Quantified Fuzzy Rules for Hierarchical Grouping of Laser Sensor Data in Intelligent Control

In complex systems it often occurs that relevant information about the system state and behavior is provided by groups of low-level variables rather than single variables. This grouping into high-level variables introduces a hierachy in the knowledge that can only be captured by means of rules involving propositions with a representation capability that is more complex than usual ones. In this paper we describe a genetic programming based approach for automated learning of Quantified Fuzzy Rules that are capable to deal with such representation capability. An application of this approach for hierarchical grouping of the distance measures provided by the laser sensors of a mobile robot (for the wall-following behaviour) is presented. Experimentation results show the control action is acceptable although no prior knowledge on the variables definition and structure was introduced in the controller. Keywords— Evolutionary algorithms, Genetic Programming, Fuzzy Quantification, Intelligent control.