In the last decades, Robot-Assisted Minimally Invasive Surgery(RAMIS) has shown its great potential of benefitting both surgeons and patients. However, most RAMIS tasks still rely on manual control, thus the main performance of a RAMIS would mostly depends on the level of the surgeon or manipulator. Since corrections and errors are inevitable in manual control, the robot path in the task would have differences from an ideal trajectory, even for robot imitation learning. In this paper, both Reinforcement Learning and Learning from Demonstration are used to generate a smooth moving trajectory without the dependency on human kinematics data. The method was trained and early validated in a simulation, Asynchronous Multi-Body Framework (AMBF). Then da Vinci Research Kit is used to validate real case performance. The results show that this path generation framework could automate repetitive surgical task.