Loading...

Proposing an Empirical Motion-Time Pattern for Human Gaze Behavior in Different Social Situations Using Deep Neural Networks

Tabatabaei Moghaddam, Ramtin | 2023

137 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 56259 (08)
  4. University: Sharif University of Technology
  5. Department: Mechanical Engineering
  6. Advisor(s): Taheri, Alireza; Meghdari, Ali
  7. Abstract:
  8. A social robot is an artificial intelligence-based system designed for humans and other robots. These robots can be used in various settings. The more closely a robot resembles and behaves like a human, the more attention and popularity it will gain. The aim of this research is to humanize the behavior patterns of the robot, especially its gaze in different social situations. The social situations considered in this study are common in human life. In these situations, the robot should make decisions based on the activities that humans are engaged in. The research consists of three parts that design neural networks using LSTM and Transformers architecture. The first part utilizes data collected from an eye-tracking device during an animation film. The second part includes not only the animation film but also human stimuli. The third part involves data obtained from virtual glasses. Finally, the designed networks are implemented on the Nao robot, and the performance of the models is evaluated on the robot. The results of the first study indicate that the designed models can predict untrained social situations with an accuracy of approximately 60% when the chance of the models to identify the correct labels is equal to 1. This prediction accuracy increases to about 80% when the chance increases to 2. The results of the second study revealed that when the model detection attempts to recognize the correct label is equal to 1, the models exhibit a gaze behavior that resembles the human gaze in social situations by about 65%. Moreover, the results of the Nao robot performance questionnaire show that 57 individuals are satisfied with the tasks performed by the robot, its intelligence, and its responsiveness to human actions. However, the robot is not perceived as a social companion comparable to a human. Lastly, the results of the third study demonstrate that when two to three people are positioned in front of the robot, it can identify the person most likely to be looked at with an accuracy of 65% in untrained social situations
  9. Keywords:
  10. Human Robot Interaction (HRI) ; Social Eye Gaze ; Non-Verbal Communication ; Virtual Reality (VR)Environment ; Social Robotics ; Motion-Time Pattern ; Deep Neural Networks

 Digital Object List

 Bookmark

...see more