Research And Implementation Of Robot Soccer Decision-Making System

Recent years saw an increase in research interests and kinds of competitions in robot soccer. It is no doubt that the robot soccer is a new interdisciplinary research area, referring to diverse fields like, robotics, intelligent control, artificial life, etc. With the more and more widespread application of robot in many areas, it is becoming more and more important to set up the organized structure of multi-robots' cooperation in a dynamic environment. Therefore, multi-robots system's cooperation architecture and entirety behavior are becoming hot research points in the artificial intelligent and robotics areas. Robot soccer system is a typical multi-agents system and multi-robots cooperative autonomous system. In fact, the system is a soccer team composed of some robot players. In the condition of complicated environment, one robot soccer team competes with another robot soccer team. The team can not only defend the opponent's attack, but also attack the opponent through organizing robot team members. So robot soccer system is a cooperation system which can complete attack and defence through robots' cooperation.Robot soccer game involves multiple robots, which is needed to collaborate in an adversarial environment to achieve specific objectives or show a teamwork behavior. In such an active and noise environment, it is a challenge problem to realize the real-time decision-making and robot real-time motion control. Compared with other multi-robots system such as transporting robots in workshop or assembly robots on assembly line, robot soccer system operates in the condition of intenseconfrontation between two robot teams, so the system is needed to have higher intelligent degree. Hence as a multi-robots cooperation system, robot soccer system provides us a standard test-bed and theoretic research model of multi-agents, so as to research the technologies of real-time inference and real-time planning of multi-robots system to complete multi-task in the complicated environment. These technologies are of important significance for human's future. According to the way of system control, robot soccer system can be divided into four subsystems: vision subsystem, decision-making sub-system, wireless communication sub-system and robot vehicle sub-system. In these four sub-systems, decision-making sub-system is used as a system controller, robot vehicle sub-system as an implement mechanism, vision sub-system as an environment monitor and wireless communication sub-system as a bridge for information transmission between decision-making sub-system and robot vehicle sub-system. Among these sub-systems, decision-making sub-system is a key sub-system, so it is the focus of attention in the dissertation. The main achievements of research work are listed as follows.1. Firstly, the relationship between robot soccer and artificial intelligent is explained, the current situation and developing direction of robot soccer are analyzed, and then the basic types of soccer robot systems and the key technologies about the global vision-based centralized control soccer robot system are introduced in detail. Furthermore the decision-making system's requirements and the architectures of the global vision-based centralized control soccer robot are discussed.2. According to the characteristics of the wheeled mobile robot's mechanism, its dynamic model is established based on kinematics analysis. Then several motion control methods about the wheeled mobile robot, such as proportion control and proportion cosine control, are discussed based on the model. The grillage parameter optimization method for robot motion control function is provided to robot motion control. This method has the advantage of high reliability which means the optimum solution can be certainly searched out, even if the objective function and constraint function are of complex forms.3. According to characteristics of soccer robot problem domains, cooperation architecture based on teamwork is provided, which can realize hierarchical taskdecomposition and team cooperation. The architecture comprises formations, roles and actions, which has capability to translate sharing-task cooperation to task decomposition and task allocation. The task decomposition is completed with defining flexible roles of specific formation. And the task allocation is completed with selecting an action of specific roles. The cooperation of home robots is achieved by switch of team formations and roles transformation. All the results of decision-making under this architecture are exhibited by executed action of robots for the aim of winning the game.4. A new path planning approach for the moving target based on genetic algorithms is presented. Firstly, potential field method and B-Spine method which are popular path planning method at present are discussed. Though these two ways are simple and easy to be implemented, they have some shortcomings such as pitfall zone and no effectiveness of obstacle avoiding in the case that obstacles move fast. A new method is provided to overcome the shortcomings mentioned above, in which the path planning approach for the moving target can be divided into two relatively simple steps. Firstly the path planning for static target on the target's path is accomplished, and then the path planning for searching out an optimum point on the target's path is completed And meanwhile two the simplified and perfect fitness functions are derived for each step respectively. The genetic algorithms are used for each step, so that the global optimum solution can be found certainly. For the path planning, adequate constraint conditions are taken into account such as robot and obstacle sizes, obstacles and target motion of and robot's motion field size. The simulation results show that the method can well ensure a moving robot navigate itself to moving target when it faces new scenario on-line. This method can also be used for automatic navigation of intelligence robot in the case that the target is moving.5. A new method of soccer robot action control based on genetic-fuzzy algorithms is put forward. A MIMO fuzzy logical controller is established and the soccer robot is controlled by a set of fuzzy rules to complete required action in real-time and dynamic environment. To establish an optimum fuzzy logical controller, genetic algorithm is used to optimize fuzzy rules and membership functions so as toincrease the fuzzy logical controller's ability to rapidly respond to the change of dynamic environment. Since the kinematics model of wheeled mobile robot is taken into account, the speeds of wheels according with the actual situation can be used as the output of fuzzy rules directly. In addition, soccer robot's action hierarchical structure is set up and some action functions are designed and realized.6. The four layers decision-making and inference model is established for decision-making system. The model is made of perspective layer, decision-making layer, task management layer and action executing layer. Furthermore, a multi-agents cooperation strategy for the soccer robot is presented based on fuzzy decision-making algorithms to complete formation determination and task allocation. Since robot soccer system is operated in real-time and dynamic environment, its operation is influenced by many factors. Therefore, it becomes very difficult to set up a precise mathematical model for the decision-making subsystem. However, due to using fuzzy decision algorithms, the precise mathematical model is not needed and the battle formation and the role of each robot can be switched dynamically depending on the current situation. Experimental result shows the strategy can well ensure the system's real-time response and the effective cooperation of each robot.7. A passing-ball route searching method of soccer robots based on rolling window is put forward. Firstly, the concept of the running circle related to the control regions of the soccer players of each team is established and the passing-ball point searching in the case of large searching space is accomplished with genetic algorithms. Hence an optimum solution of the each step about passing-ball route can be found. Secondly, the whole passing-ball route from start point to the end point is searched out in the real-time dynamic environment based on rolling window. The method can be used in the unpredictable environment information. Experimental results in simulation and global vision-based centralized control soccer robot system prove the method is correct and feasible.8. A hierarchical decision making software architecture of perspective layer/ decision-making layer/ task management layer/ action executing layer is presented for soccer robot decision-making system software's design and development. The developed software, meeting to system real-time requirement, has been used forrobot soccer competition in China and the world, and the satisfied result has been achieved.