The Research Of RoboCup3D Simulation Robot

The Robot World Cup football match is referred to as RoboCup, the competition can promote the development of artificial intelligence, robotics and other related disciplines. RoboCup3D simulation game is similar to the real human soccer match, compared to the medium-sized entities group of RoboCup, it can break through the constraints of hardware and achieve some functions that medium-sized group can’t have. Through the design of robot architecture, perception module, information processing, decision-making, skills and so on, it can cooperate with each other in a complex environment, and ultimately win the game. The research results of the competition can be applied to many areas, such as space exploration, industrial and other smart fields.The first international competition of RoboCup3D is held in Lisbon, Portugal in2004.In home, the first RoboCup3D simulation competition is established in Changzhou, Jiangsu2005in where China Robot Competition is held. During the development in recent years, the simulation robot in the game has experienced a change from the spherical robot to humanoid emulation robot, now the development of domestic RoboCup3D simulation game is still in its infancy, it has not many race teams, the research of team strategy and players advanced skills is still immature, now the various colleges and universities are committed to the research of game environment and players’basic skills. In this paper, we research the simulation robot in RoboCup3D simulation games, the main contents are as follows:(1) The architecture design of RoboCup3D simulation robot.Through analyze the advantages and disadvantages of agent’s architectures commonly used, combine with the characteristics of RoboCup3D simulation environment, we choose the mixture model based on layer structure to design TAIJI3D simulation robot’s architectures, divide it into four levels:interface layer, world model, decision and technical layer.(2) The World Model of RoboCup3D simulation robot It has proposes an observation weighted fusion Kalman filter approach to realize robot self-location, and on this basic to locate other objects in the field, thereby update the world model of robot. The simulation experience result shows that the method greatly improves the accuracy of robot self-location, thus obtain a relatively accurate world model. Also research the prediction of world model.(3) Research the intercept and shoot of RoboCup3D simulation robot’s skill layerMaking a detailed study on intercept and shoot which are two important skills in simulation robot’s skill layer, dividing intercept to close intercept and long-range intercept to discuss, we use BP neural network to train long-range interception. In the case can’t shoot directly using A*search algorithm to find the shot path.Finally, summary the dissertation, it makes out the main research results and the problems which need to research in the future.