Multi-Agent Research & Application

This paper research Multi-Agent Systems, which comes to light, Multi-Agent Systems & distributing artificial intelligence have become artificial intelligence research's emphases.The essential problem of Multi-Agent is correspondence between Agents, which can be fractionize as the design of single Agent, Multi-Agent's structure, Multi-Agent's collaboration & communication, automatic illation, layout, machine learning & KA, cognition modeling, system zoology, evolution, and so on.RoboCup is an image representation of Multi-Agent Systems, it's a lagre match & learning activity for improving education and research's level in correlative domain. RoboCup is a good flat roof for Multi-Agent Systems research. Various fields knowledges make up of a success RoboCup team. Reasonable model-structure and correspondency & collaboration are the key to prevail in the RoboCup matches, so they are the important problem in the Multi-Agent researchs.We have done some researches and practices in the Multi-Agent knowledge domain by the RoboCup flat roof, including: in the team technique, combined the UVA's bottom techniques, we have extent and amended some middle-level and high-level techniques, such as: using a fuzzy reinforcement learning method to amend kick technique, so that Agent manage to obtain anticipant ball's velocity and direction in shorter time; using genetic algorithm in intercepting ball, so that Agent intercept ball more efficiently; amending Agent's single tactic-using influence map for former field Agent to select action; designing Multi-Agent's team tactic-using three levels collaboration model,importing static & dynamic lineup concept,exerting role to Agent that Agents try to complete their own task then assist each others if necessary; advancing my learning brother's team include importing alarm into four levels structure model, amending shooting and passing technique; adjusting the codes and exert line course mechanism so that the codes are more efficient.