Intelligent Agent Research And Practice

The Robot Soccer has been became the research center of many Artificial Intelligence researchers. As the ideal testing platform of Multi-Agent Systems, it is involved in many technical domains. At now, RoboCup is the most important Robot Soccer Game in the world. The author has designed and realized a RoboCup simulation league based on the theory of Agent and Multi-Agent Systems.There are lots of considerations in the process of a simulation league designing: the communication between clients and server, message parsing, the architecture of Agent, the training of Agent's skills, the collaboration among Agents, the confrontation with opponents and the strategy of whole team, etc. In the process, we focused our mind on the following aspects:First, the author designed a layered mixed Agent architecture suite for RoboCup. Our mixed architecture had four layers: interface layer, parser layer, decision layer and layout layer. With this architecture, the Agent can not only response to some emergencies but also make layouts for near future.Second, the author introduced a new method based on decision tree learning to the training of Agent's skills. Besides estimating the action performace efficiency, the Agent can choose a better action according to the situation on the field when our method is applied.Third, the author introduced a local strategy of collaboration based on tactic, which realized the consciously collaboration of Agents and improved the cooperation and confrontation ability of the league.Last, a new method based on Analytic Hierarchy Process was used to choose heterogeneous players under the consideration that different players have different duties.