PURPOSE: To automatically provide a stable control system even if a designer does not accurately know various parameters of the mass, etc., of a controlled system (plant).

CONSTITUTION: An adaptive controller is provided with a controller which controls the internal variable parameter to show the action of a desirable controlled system (plant) as a normative model and to secure the coincidence between a real plant action and the action of a normative model. The controller consists of a neural network and uses the signal (Δn) obtained by multiplying an error (e) between the actions of the real plant and the normative model by the proper gain (K) as a teacher signal. A neural network which learnts the forward model of the plant is provided and the error (e) is inputted to the output side of the neural network. Then the error (e) is adversely transmitted based on a generalized delta rule. Thus the input error (Δu) of the plant is obtained and used as the teacher signal of a neural network controller.