Online estimation of arriving time for robot to soccer ball in RoboCup Soccer using PSO-SVR

Abstract

Automation using robots is applied widely in industry and recently, it is used to more and more complicated works. In order to apply it to complicated works, multi-robot operation is necessary and to multi-robot operation, coordination of robot movements is inevitable. This paper proposes an algorithm to estimate arriving time for a robot to soccer ball in RoboCup Soccer Simulation 2D League. In the league, the player robots are operated by league server and the characteristics of robots are unknown to playing team controller and also the time to get soccer ball is unknown in advance. Hence, the controller needs to estimate the catching time to decide which robot gets the ball fastest and should get ball. The estimation is based on a prediction model obtained by Support Vector Regression (SVR), which is one of machine learning methods. Moreover, the SVR method is optimized by Particle Swarm Optimization (PSO) method to increase predicting accuracy. The method developed in this paper has possibility to be able to apply other multi-robot operations and other prediction problems. The effectiveness of the proposed method is shown by numerical simulation.