Online stability and direction fall detection for robotic soccer players using feedback from server

Abstract

Nowadays, robot stability is one of the most important challenges of 3D Soccer Simulation League in the science of robotics. Loss of balance and falling on the ground by colliding two robot players, is one of the most important topics in the team-collaboration-bots, and this can beneficial for the opposing team. In this work we present a new approach for detecting robot stability, based on information derived from the 3D soccer simulation server(rcssserver3D). In this article, two data sets using the positions of the joints of the robot legs are generated. One of the dataset is based on the sustainability of the robot and the other is based on the falling direction (left and right). A model based on SVM learning algorithms and neural network, are employed. We add this model to a 3D soccer simulation platform, in order to recognize the stability and falling or not falling situation(direction) online. By this model stability of each robot is notified to his teammate and consequently adequate information is provided for best performance. In the SVM model, with using kernel RBF together with optimizing parameters was used to get optimal performance. The significance of this approach is due to its applicability when number of samples is low and is able to appropriate action.