REAL-TIME LEARNING OF NEURAL NETWORKS AND ITS APPLICATION TO THE PREDICTION OF OPPONENT MOVEMENT IN THE ROBOCODE ENVIROMENT

Abstract

The paper deals with a general information about the Robocode environment, and the application of artificial neural networks to the prediction of enemy's robot movement. This paper proposes to use a neural network with time delays as a predictor, and the selection of the best prediction network is carried out based on different network structures. The paper includes also a comparative study of Back-Propagation (BP), Conjugate Gradient (CG) and LevenbergMarquardt (LM) training algorithms.