Abstract: This paper discusses the Developmental learning algorithm and its application to technical action learning of soccer robot. According to the advantages of the developmental learning, we choose the proper reinforcement learning algorithm, and apply it to design technical actions learning of soccer robot without any prior knowledge and environment model involved. It can autonomously improve its behavior policy with the knowledge obtained by continuously interacting with the environment．Finally, the experiment results are also shown to prove the presented method is effective and superior, and it can meet the demands of the high level robot soccer match．