Loading...

Learning Interactive Skills of NAO Robot through Imitation Learning from Observation

Alizadeh Kolagar, Adel | 2022

267 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 55202 (08)
  4. University: Sharif University of Technolog
  5. Department: Mechanical Engineering
  6. Advisor(s): Meghdari, Ali; Taheri, Alireza
  7. Abstract:
  8. Using learning from demonstration approaches is one of the ways to teach a robot how to perform different skills. In classic methods of learning from demonstration, the subject should wear some sensors and devices which is and expensive way. But, with the progresses that have been made in the context of computer vision, it is possible now to achieve to the same results with less cost. Learning from observation is the approach that would be used in this project to do that. By using this method, when robot watches a video, it tries to understand the behavior of the subject and behave in the same way.The purpose of the project is that the robot learns some social (interactive) skills through watching videos and use them while interacting with humans. To achieve this goal, 2 steps should be done. First, imitation leaning of the robot and next social interaction with a human. For the first step, a pose estimation algorithm is used to extract the movement information of the subject in the video. Then, the extracted data would be transferred from human space to robot space, as a result, the robot would be able to learn how to perform the skills from them. For the social interaction step, the robot should be able to understand the skill being performed by the subject. To do so, an action recognition algorithm is used on the extracted data from pose estimation step. After recognizing the action of the subject, robot performs the correct response from the learned skills according to a predefined scenario.The expected result in this project is that the action recognition and performing the response would be done correctly and in a real-time manner. In case the expected result be achieved, this approach can be used to teach other things to the robot. Due to high speed of the training process, large amount of training data and no need for an expert, this method can be so effective
  9. Keywords:
  10. Human Robot Interaction (HRI) ; Action Recognition ; Robot Control ; Learning from Demonstration (LfD) ; Postural Estimation ; NAO Robot

 Digital Object List

 Bookmark