Design of fuzzy controllers with local response neurons

The class of artificial neural networks made up of nodes that have a localised response in input space have functional similarities with fuzzy logic controllers. The localised response nodes can be interpreted as representing membership functions. Each node represents a fuzzy rule for a control action and the network interpolates between the fuzzy rules. We illustrate the process of building a fuzzy controller for the cart-pole experiment using a local response neural net based on clusters of neurones with sigmoidal transfer functions.<>