Abstract
A simplified fuzzy reasoning method and a iterative learning algorithm based on the steepest gradient method are proposed. The inverse dynamics model is identified by the iterative learning algorithm. By the simplification of the reasoning, a mathematical models of fuzzy reasoning can be represented by a multi-linear function. The efficiency of the iterative learning algorithm is proved in cases of parity check digits and higher order functions. The convergence of the algorithm is discussed. And, an advanced algorithm using stepwise refining of the fuzzy partition is proposed. As an example of learning control, a computer simulation of robotic manipulator is presented and compared with the ones by neural network models.