Research On Goalie Control For RoboCup Middle-Size League Robot Soccer System

Robot soccer competition system, as a major challenge in the field of artificial intelligence, is a hot topic of current research. It incorporates the multidisciplinary research achievements, and provides a standard experimental platform for the research of robot technology and multi-agent system. Middle-size league soccer robots are completely autonomous, communication through the wireless network, and each teammate coordinates between each other to complete the race task. As one of the important members, goalkeeper takes a team’s defensive task, and its control performance plays a very important role in the result of a competition. In this paper, take the Middle-size league soccer robot goalkeeper as the research object, based on collecting and analyzing the related literature at home and abroad, combined with the specific situation of robot soccer competition, we conduct a study about the behavior of the goalkeeper, including football position prediction, motion control, defensive strategies. The main contents include:1. Football position prediction.Football is a key target for the goalkeeper. By predicting the motion position of football, the goalkeeper can timely take some sort of defensive strategy to prevent the football into our goal. Taking into account the dynamic and real-time in the actual robot football game, an algorithm based on calculation of strong tracking and H∞filter is proposed to predict the position of football. By introducing a fading factor into H∞filter algorithm, it improves the prediction accuracy. Experiment results show that the algorithm can track football mutation quickly and predict the football position more accurately.2. Goalkeeper motion control.Goalkeeper executes the corresponding motion through controlling the movement, and then realizes motion control. Based on the relevant provisions of coordinate system in robot soccer competition system and robot kinematics model, a control algorithm for goalkeeper’s motion based on improved PID is proposed. Through the analysis of the implementation of each motion, it gives the corresponding control. Experiment results on following football show the good performance of the control algorithm. It can satisfy the competition requirements.3. Goalkeeper defensive strategy.In order to make goalkeeper intelligent in defense decision-making ability and execute the corresponding defensive motion according to the change of competition state, on the basis of analysis of reinforcement learning method, a reinforcement learning algorithm based on Elman network is proposed to realize defensive strategy of goalkeeper. Through introducing Elman network into Q-learning process, it improves the performance of learning. Experiment results show that the algorithm can effectively improve the defensive ability of goalkeeper and realize the independent defense decision-making of goalkeeper.Finally, the paper summarized on the main research results, and points out the subjects to further research and improvement in the goalkeeper defensive control in the future.