Abstract: This paper presents a hybrid intelligent learning algorithm based on multi-agent, and it combines individual learning with group learning effectively. The specific application and experimental results in RoboCup simulation system show the feasibility and effectiveness of this algorithm.