Behavior learning to predict using neural networks (NN):: Towards a fast, cooperative and adversarial robot team (RoboCup)

To build a fast, cooperative and adversarial robot team (RoboCup), prediction behaviors became necessary. Indeed, such behaviors can be obtained by learning from past experience bringing the overall team behavior near to that of human team one in the learning, adaptation, generalization, and prediction. In this paper, a behavior learning to predict using Neural Networks (NN) is developed to enhance the behavior of GMD mobile robots. In fact, the suggested NN called NN-Prediction learns to predict successfulness of the elementary behavior "Kick" the ball towards the goal in order to act as consequence. The training is carried out by supervised Gradient Back-Propagation (GBP) learning paradigm. This NN-Prediction has been specified on the Dual Dynamics Designer (DD-Designer), to be thereafter implemented and tested on both the Dual Dynamics Simulator (DDSim) and GMD mobile robots, and analyzed on the Real-Time Trace Tool (beTee). NN-prediction demonstrated, during the e World Championships RoboCup'2000, cooperative and adversarial behaviors especially face to situations where the successfulness of "Kick" is not guaranteed. Then, a discussion is given dealing with the suggested prediction behavior and how it relates to some other works.