الفهرس Only 14 pages are availabe for public view
 |  | from | 95 |  |  |
| | | from | 95 | | |
Abstract
Many life applications are extremely depending on using the robots, thus humans are seeking to develop efficient robots. Developing such robots necessitates programming the robot, thus the machine learning approaches are employed to program the robot, and to achieve objective configuration by obtaining additional information. Programming the robot can be applied by demonstration such that the skills are transferred to robots by providing examples of the required behavior. This programming technique is analogous to the way humans learn new skills (i.e. demonstrations by a teacher). The key motive of using this approach is the elimination of the need for extensive technical knowledge in order to program the robots. Learning a skill at a trajectory level involves modeling set of demonstrated trajectories and retrieving a generalized representation of the set suitable for reproduction by a robot learner. The objective of this research is to develop efficient and cost-effective technique for trajectory learning. Trajectory learning model includes three main phases: preprocessing, clustering, and modeling. However, the research in the trajectory learning area concentrated on the modeling phase; we are concentrating on the first and the second phases in this thesis. For the first phase (i.e. preprocessing) the most recent research has only handled the NANs fields included in the raw data and applied smoothing for the data. Then, these data are used for finding set of initial key points representing each demonstration. Applying the clustering and modeling on the raw data may cause distortion in the learning outcome. Therefore, a reconstruction for the raw data in the preprocessing step is necessarily to handle this distortion. In this thesis, we propose new schemes for generating a generalized trajectory by employing set of demonstrated trajectories. Two schemes have been proposed for the purpose of preprocessing the data. In the first scheme, Principal Component Analysis has been used for reconstructing the data set. In the second scheme, Posterior Hidden Markov Model State Distribution has been used for the same purpose of reconstructing the data set.