This paper presents a method of implementing distributed planning in partial-observable, low communication bandwidth environments such as RoboCup simulation games, namely, the method of global planning from local perspective. By concentrating planning on comprehending and maximizing global utility, the method solved the problem that individual effort might diverge against team performance even in cooperative agent groups. The homogeny of the agent decision architecture and the internal goal helps to create privities among agents, thus make the method applicable. This paper also introduces the application of this method in the defense system of Tsinghuaolus, the champion of RoboCup 2001 Robot Soccer World Cup Simulation League.