Collaborative optimization (CO) is easily influenced by the initial points, whose stability and convergence are poor. To solve this problem, the coupling consistency degree was defined to measure the consistency of the variables in different disciplines. The asymptotic relaxation based collaborative optimization method (COAR) was proposed，whose calculation process was divided into two phases. In the first phase, the relaxation factor of the system was computed based on the set coupling consistency degree, and the approximate global optimal solutions was solved, which were adopted as the initial points in the second phase. In the second phase, the relaxation factor was fixed again at a minimum according to the precision demand of the coupling consistency, and the final solutions were obtained. The standard numeral example of the reducer shows that the proposed method can greatly reduce the sensitivity to the initial points and effectively improve the convergence and the stability of CO.