Planning robotic radiosurgery treatments for multiple (n > 3) metastatic brain lesions is challenging due to the need of satisfying a large number of dose-volume constraints and the requirement of prescribing different dose levels to individual targets. In this study, we developed a sequential two-step optimization technique to improve the planning quality of such treatments. In contrast to the conventional approach of where all targets are simultaneously planned, we have developed a two-step optimization method. In this method, the first step was to create treatment plans for individual targets. In the second step, the 3D dose matrices associated with each plan were exported to Dicom-RT digital files and subsequently optimized. For the optimization, a singular-value-decomposition (SVD) algorithm was implemented to minimize the dose interferences among different targets. Finally, we compared the optimized treatment plans with the treatment plans created using the conventional method to determine the effectiveness of the new method. Large improvements in target dose distributions as well as normal brain sparing were found for the two-step optimization treatment plans as compared with the conventional treatment plans. The two-step optimization significantly lowered the volume of normal brain receiving relatively low doses. For example, the normal brain volume receiving 12-Gy was reduced by averaged 42% (range 34%-47%) with the two-step optimization. Such improvements generally enlarged with increasing number of targets being treated regardless of target sizes. Of note, normal brain dose was found to increase non-linearly with increasing number of targets. In summary, a two-step optimization technique is demonstrated to significantly improve the treatment plan quality as well as reduce the planning effort for multi-target robotic radiosurgery.