In the cooperative navigation algorithm for multiple Autonomous Underwater Vehicles (AUVs) with a single leader, the model of the system is nonlinear. The Extended Kalman Filter (EKF), which is directed against the nonlinear system, is one of the most influential techniques. However, the performance of EKF critically depends on a large number of modeling parameters which can be very difficult to obtain, and are often set by manual tweaking and at a great cost. In this paper, a method for automatically learning the noise covariance of a Kalman filter is applied, and the simulation result shows that this algorithm fully automatically and quickly outputs the noise covariance, which improves the navigation accuracy of the cooperative navigation system.