A learning scheme for open-loop and closed-loop control

Abstract

The problem of control design for systems to perform repetitive tracking is considered. The control structure combines open-loop control and closed-loop control. The specific organization is designed on the philosophy that open-loop control should be used to take full advantage of a priori knowledge and feedback control should be used for regulation against modeling uncertainties and disturbances. With this control structure learning takes place in the open-loop controller. In each iteration the share of the open-loop control in the total control input increases, while that of the closed-loop control decreases. Convergence theorems are established as design guidelines for learning algorithms. Numerical examples are given.<>