Specialization versus generalization in neural network learning for ballistic interception movement

This paper presents the use of an artificial neural network (ANN) in learning the features of a dynamic system, in the special case of implementing the controller to launch objects under ballistic movement. We make considerations about the generalization capacity both for live beings and for the network. We also present the network topology and configuration, the learning technique and the precision obtained. The importance of the activation function choice in learning some critical points is shown, as well as considerations on the distance between examples are presented.