Craniotomy is a prerequisite for many neuro-surgeries such as intracranial tumor resection and decompression of cerebral hemorrhage. However, existing manual craniotomy methods are time-consuming and labor-intensive, presenting low efficiency and security. Robots have shown great potential for safe, precise and efficient craniotomy by introducing accurate positioning and stable motion control. We proposed a method that automatically computes operation path for robotic craniotomy and adapt to the irregular geometric change of the skull, based on the surgeon’s surgical plan from computed tomography (CT) images. The drilling path is generated from intersection between the skull entity and surgeon-input drilling intents. A virtual-center (VC) method is developed to adaptively compute an initial milling path through the generated drilling path, which is then improved under both clinical and skull-cutting constrains. The results show that our method works effectively to generate the operation path adapting to the unstructured skull. This research highlights the potential for optimizing current craniotomy procedures by employing the robot, laying the foundation for future autonomous robot-based craniotomy.