On the stability and convergence properties of the fuzzy learning algorithm for robotic manipulators

Investigates the stability and convergence properties of the fuzzy learning algorithm which was the subject of the authors´ earlier research. The authors show that the algorithm lends itself more fully to analysis than traditional intelligent control techniques. The authors´ analysis begins with conditions under which a fuzzy regression problem will have a solution, and then considering a bound on the fuzzy regression cost function and showing that this bound leads to a bounded error norm, which implies stability, and also convergence if the bound can be shown to be zero. The authors conclude that they can control the stability properties of the algorithm through the selection of the fuzzy regression parameters, and that in general, the less fuzzy the model learned by fuzzy regression, the closer the algorithm will be to providing overall trajectory convergence