Toolpath generation and optimization is considered as a challenging problem in the minimally invasive orthopedic surgery with a milling robot. The objective of this paper is to minimize the collision of the cutting tool with the soft tissues. A novel approach of toolpath generation and optimization is proposed. A redundant axis is implemented to avoid the collision in the robot. Some important components are modeled based on the physical requirements. A geometric optimization approach based on the model is proposed to improve the toolpath. Case studies show the validity of this approach. Software is developed for this application and the effectiveness is evaluated with a cadaveric bone.