In the process of hot stamping of steel sheet, there may appear crack and wrinkling defects. Therefore, in the early stage of die design, it is necessary to evaluate the reliability of its quality and optimize the original design. Considering that there are both random and fuzzy factors in the stamping process, the entropy invariance principle is used to transform the fuzzy factors to the random ones, and the equivalent hybrid reliability model of anti-defect is established. Then, the reliability index is taken as constraints in robust optimization. Finally, the stamping process of bulge formed joint is taken as an example, the maximum thinning rate is taken as objective, and the reliability of anti-crack and anti-wrinkle is taken as constraints in reliability-based robust optimization (RRDO). The results show that the standard deviation obtained by robust optimization is 77.5% less than the standard deviation obtained by deterministic optimization and the reliability obtained by RRDO is higher. Compared with the experiment result, the result obtained by RRDO is more in line with engineering practice.