Abstract: It is difficult for the existing models of cooperative coevolution to effectively produce collective complex adaptive cooperation behaviors because of their deficiency in solving problems with interaction between subsystems.In order to solve this problem,a new mechanism of cooperative coevolution with local interaction for collective complex cooperation behaviors is proposed based on system theory and nonlinear science theory.This mechanism leads the cooperative coevolution to the correct direction by combining global elicitation information with local elicitation information detected by local interaction.The algorithm analysis and simulation experiment of cooperative multi-robot box-pushing show that this mechanism can effectively overcome the deficiency of the existing cooperative coevolution models,and can make the cooperation behaviors with complex correlation rapidly evolve to the global optimum solution.