The existing research of the industrial robotic motion planning mainly focuses on the trajectory planning and the path planning, which aim at obtaining a minimum-time trajectory and finding an obstacle-free path respectively. In the trajectory planning, the jerks of robotic joints are usually not considered, so the smoothness of the robotic motion cannot be ensured. In the path planning, the complex spatial curves generally cannot be tracked easily, and it needs the fussy teach-and-playback operation. In this paper, based on the different constraint characteristic of the variable motion phases in the practical robotic task, the complicated task is decomposed into two kinds of sub processes, which are the free motion process and the constrained motion process. In the free motion process, the kinematic models of quasi trapezoidal waveform and quasi triangular waveform are proposed with the dynamic limits of maximum velocities, accelerations and jerks of robotic joints, so the minimum-time trajectory can be obtained with the motion smoothness. In the constrained motion process, the mathematical presentation of the task paths is extracted from the CAD models of the workpieces, so the complex spatial curves can be tracked autonomously without much teaching operation. In addition to the theoretical research, a robotic virtual prototyping system is also developed for the industrial arc welding production. The proposed research ensures the robotic motion smoothness, enhances the ability for tracking the complex spatial curves, and improves the quality and efficiency of the robotic motion planning.