부산대학교 도서관

Many applications in our everyday living are totally depending on using the robots. So that, the need for having smart and more productive robots is increasing. Developing such robots necessitates the programming of the robot. Hence, the machine learning approaches are widely employed to accomplish this objective successfully. Programming the robot can be applied by demonstration such that the skills are transferred to robots through supplying examples of the desired movement. Multiple trajectories are observed and modelled in order to learn new skills by obtaining a general trajectory of the appropriate set. Traditionally, output distortion may be occurred by modeling the input trajectories as discrete data. In order to overcome such distortions, a preprocessing stage of the raw data is needed. Recently, preprocessing trajectory learning techniques have been proposed. It is found that these techniques can increase the accuracy of the generalized trajectory. In this paper, we present a comparative study for these preprocessing trajectory learning techniques in terms of accuracy and the computational cost. It is shown that adding the preprocessing stage to the learning models in generating a new trajectory can increase the accuracy at reduced computational costs in comparisons to related work.