Progressive optimization of a fuzzy inference system

Describes an automatically online tuned fuzzy navigation system for an autonomous robot using a modular structure to generate the angular speed as a function of the sensor data. The goal is to obtain a reactive behavior, such as wall-following, with the adaptivity necessary for coping with large modifications in the physical characteristics of the robot. For this behavior, the building of the navigation controller is done entirely online by the optimization of a zero-order Takagi-Sugeno fuzzy inference system (FIS) by a backpropagation-like algorithm. It is used to minimize a cost function that is made up of a quadratic error term and a weight decay term that prevents excessive growth of the parameters of the consequent part. The procedure is performed entirely online, but in two steps. The first one is done on a miniature robot or on a dedicated simulator. Then the obtained controller is transferred to the real robot and a further optimization step is performed. At the end of the procedure, it is possible to extract knowledge by interpreting the result parameters in a symbolic form. One can notice that the two tables deduced for the miniature robot and for the real robot are very close with respect to their linguistic concepts. Moreover, these two automatically extracted rule tables are quite close to those written empirically, but we can observe that some human expertise rules work wrongly because the expert doesn't expect a particular situation. In fact, the main advantage of this procedure is the optimization of the controller with respect to the actual characteristics of the robot. That means that, for example, the rough manual tuning of the global gain acting on the width of the universe of discourse is replaced by a fine local automatic tuning, and this improves the performance very significantly. This method is simple, economical and safe, since it is done on a miniature robot. It leads to a very quick and efficient optimization technique.