A reference model unit calculates a model position, with which a model of a controlled object follows a position command, and a model torque for the model of the controlled object to operate to coincide with the model position. A gain changing unit changes, during the operation of the controlled object, at least a value of one control gain of first-order and second-order control gains used for calculation of a variable compensation value output to an integral compensator by a variable-compensation calculating unit based on at least one of the model position, a position detection value, and a torque command output to the controlled object by a torque adder, and reflects the value on calculation of the variable compensation value.