Aircraft design is a complex multidisciplinary process which requires collaboration of multiple and diverse teams of engineers from different disciplines or subsystems. The design and optimization process generally involves large number of design variables and coupling variables among disciplines or subsystems. 1, 2 In a complex and coupled system like aircraft, there is a huge computational burden to run a multidisciplinary analysis (MDA) for one design because it involves several iterations to converge to a design which satisfies interdisciplinary compatibility. Therefore, the traditional methods3 such as All-In-One (AIO) optimization method, which treats the entire MDA process as a black box, become computationally expensive for a design and optimization process. Also, the preliminary and detail design phase of design process involving high fidelity tools and disciplinary experts, leading to intensive computational expense for a disciplinary analysis and limited co-ordination between various disciplines. To address these issues, various multidisciplinary optimization (MDO) methods have been developed4 for decomposed systems to maintain disciplinary autonomy and distributed sub-optimization, under a decentralized multidisciplinary design environment. One such method is collaborative optimization (CO). 5 CO decomposes and decouples various disciplines so that each discipline can simultaneously and autonomously carry out its design process. The coupling between the disciplines are handled by introducing interdisciplinary constraints on auxiliary design variables and shared variables. In other words, the CO method has a system level coordinator which handles the system level optimization in addition to handling the interaction and coordination between subsystems. The subsystems carry out local optimization concurrently to meet targets for coupling and shared variables specified by system coordinator. CO was originally developed for deterministic optimization. However, for many practical problems, there are uncertainties at the subsystem level. The uncertainties can arise from prediction errors induced by assumptions and simplifications of computational models in a disciplinary analysis, operating conditions, coupled and shared variables, manufacturing tolerances, etc. 6 The deterministic CO framework has been extended for robust design and subsequently, a number of robust CO (RCO) approaches have been developed for uncertainty management strategies. Du and Chen7 describe two approaches—the extreme condition approach and statistical approach to propagate uncertainties. The extreme condition approach has been used for calculating the intervals of the output. The statistical approach has been used for calculating the statistical distribution. Wang et. al. 8 developed a novel RCO method based on dual response surfaces. In this method, two response surfaces replace the mean and standard deviation of state variables and performance metrics. However, this method requires accurate response surfaces, particularly for nonlinear responses. This in turn requires analysis on large sample points. Another RCO method based on implicit uncertainty propagation (IUP) 9 has been developed in which state variables are considered as auxiliary variables and their intervals are estimated using global sensitivity equation (GSE) in an IUP module. However, IUP relies on first order Taylor series approximation of state variables, which can be inaccurate for state variables containing non-linearity or large variations. Xiong et al developed a moment matching RCO (MM-RCO) 10 which overcomes the issues of earlier RCO methods by …