Multiple Objective Optimization of Fuzzy Rules for Obstacles Avoiding by an Evolutionary Algorithm

We focus our attention on the classical problem of a simulation of obstacles avoiding. We tempt for this to make raising lists of fuzzy rules by an original evolutionary algorithm. We look for an emergence of such rules system for a robot which is able to change its direction and its speed, without precise constraints. It is really not a well-defined problem, because when there is a target (some zona to reach) or if there is a constant speed, genetic algorithms have already given good solutions. Here, we would wish to see a birth of a good road-holding comportment, that is to say to slow down in the curves and to accelerate in the straight lines. In the same time, we want to get the best response to the obstacles avoiding problem, namely solutions which are satisfying in every occasion. We are face to the well-known dilemma of learning: a good solution in a precise playground will be a set of particular rules, solutions given by randomly circuits will not be comparable themselves and it is difficult to evaluate a solution on a big number of different playgrounds. On the other hand, we wish short and readable solutions, so, as every concrete problem we have several objectives to optimize. I THE FUZZY CONTROLER AND ITS ENCODING In the field of problems raised by the tuning of a fuzzy controller, the first is the choice of inputs. So, if a robot is able to evaluate the distances from it to the obstacles, the rules may be very differen