The multiobjective robust collaborative optimization framework consists of optimization both at the system and autonomous subsystem levels. Linear physical programming is used in the system level optimization, which avoids the difficulty in choosing the multidimensional Pareto set. The non-dominated sorting genetic algorithm (NSGA-II) is used in the subsystem optimization with physical objectives. The interdisciplinary incompatibility function and physical objectives have different priority levels. At the first priority level, the best individual should be in the feasible region of the subsystem. At the second priority level, the interdisciplinary incompatibility function of the best individual should be no more than the feasibility threshold. The physical objectives are improved after the achievement of the above levels. A method for producing initial population with feasibility and diversity is proposed to improve the calculation efficiency and accuracy of the subsystem optimization at the first priority level. A method for setting dynamic feasibility threshold is proposed for the non-dominated sorting to help the physical objectives to obtain better solutions at the second priority level. Finally, the results of the speed reducer show that the presented method is efficient.