Collaborative optimization (CO) is the most widely used Multidisciplinary design optimization (MDO) method for the design of complex engineering system. But some serious computational difficulties are found in its application. Reasons that cause computational difficulties in original CO were analyzed and a new improved collaborative optimization method (ICO) was presented. The L1 norm was used to improve subsystem consistency constraint and to avoid discontinuities in subsystem object function derivatives. Penalty function was added to system-level object function to convert constrained optimization into unconstrained optimization. A quick-start strategy was used to make the best use of optimal solution of system-level optimization in subsystem-level optimization. Experimental results show that the robustness, reliability and computing efficiency of ICO are higher than CO.