The neural network model RuleNet and its application to mobile robot navigation

In this paper the neural network models RuleNet and its extension, Fuzzy RuleNet, are described in detail. RuleNet is a feedforward network model with a supervised learning algorithm, a dynamic architecture and discrete outputs. The main characteristics of RuleNet are its efficient learning and propagation algorithms and the possibility to translate symbolic knowledge into the network and vice versa without loss of information. Fuzzy RuleNet is an extension to RuleNet with Fuzzy Logic. The main application area of this neuro-fuzzy model is fuzzy classification. An important characteristic of Fuzzy RuleNet is the possibility of knowledge transfer into and from the network without loss of information, therefore it can also be used for the generation of fuzzy systems (i.e. fuzzy rules and the corresponding membership functions). RuleNet and Fuzzy RuleNet have been applied to a hierarchic behavior based navigation system for mobile robots. As a first step, a wall following behavior has been implemented utilizing these network models. The achieved results in the simulation environment as well as on a mobile robot experimental platform are very encouraging.