Albert Chen
Research Mentor(s): Wonse Jo
Mentor Department: Robotics
Authors:
Session: Session 1 (9:00am – 9:50am)
Presentation Type: Poster 37
Abstract
Human visual perspective-taking (VPT) is a critical cognitive skill that enables individuals to understand the visual experiences of others from their viewpoint. In the context of Human-Robot Interaction (HRI), integrating the ability of human’s VPT is essential for creating robots that can effectively collaborate with humans by understanding their intentions from their perspective. However, there are still limitations to extracting human visual perspective-taking from robot sensors. Therefore, this project proposes a reliable real-time algorithm for extracting human VPT from robot sensors. By utilizing the robot’s visual simultaneous localization and mapping (V-SLAM) and the extracted human behavioral features from robot’s head cameras, we aim to equip robots with the ability to understand human VPT, identify where people are looking, what objects they are interested in, and whether those objects are related to ongoing interactions with the robots. Finally, we compare the group data (e.g., human agent’s viewpoint in simulation and eye-tracking sensors’ global viewpoint in pilot tests) with the extracted VPT from our algorithm. In the future, we plan on testing the algorithm using real sensors and refining the algorithm. By doing so, our VPT algorithm will open the door to the development of adding human features to robots, further improving human-robot interactions.
