PURPOSE: To enable the robust quadratic optimization while considering both a model error and the influence of random disturbance by performing the optimization of a selective partial input consisting of one part of a control input to a dynamic characteristic model based on the state of a controlled system.

CONSTITUTION: First of all, a controlled system 10 such as the dynamic characteristic model of a robot, for example, is prepared by a control model generating means 11. This model is reflected with the deviation of a true value in a coefficient matrix or noise. Since the partial input correcting the selective partial input introduced for the documents of a control system design theory is considered for this model, the generated dynamic characteristic model can be handled as a standard probability system model and since the partial input is optimized by applying the robust quadratic optimization rules, the optimum partial input can be provided. Based on this optimum partial input and the output of the control model generating means 11, the control of the controlled system 10 is performed.