This research describes a dynamic role allocation and fast formation positioning method in the multi-agent robot soccer domain. The optimal role assignment is a well-known problem from operations research, and the problem-solving techniques include the simplex method and the Hungarian method. The goal of those methods is to calculate the running time of positioning and to minimize the sum of all costs. However, the computing concept is different from the practical positioning, because the consuming time of practical positioning is determined by the time it takes for the last agent to reach its target position. In the robot soccer, both of role allocation and formation positioning must be performed in real time, so that robots can be effective in a dynamic environment. Therefore, we propose a new method with dynamic role assignment mechanism to find a near-optimal solution. The proposed method produces a set of corresponding combination quickly, and it is an acceptable approximate solution for the problem. We also provide time complexity analysis of our algorithms in this study, and the evaluation indicates that our method is much faster to complete the role assignment than dynamic programming approaches. Our experiment shows that the output of one-to-one mapping between agents and roles through the initial role allocation and re-allocation. Furthermore, in a robotic soccer game the number of available robots varies as a result of several situations such as hardware or software failures and penalties. Hence, we also use different priorities for the different roles and positions, so that the most important ones will always be played to fulfill the obligations.
展开▼