Online and Unsupervised Face Recognition for Humanoid Robot: Toward Relationship with People

摘要

The Problem: The ability to distinguish among different individuals is crucial for all social animals. Observations of behavior within dolphin male coalitions indicate that dolphins are able to not only recognize friends, but also keep track of their previous behavior to predict future action. This work addresses the questions of how one may implement such social competence in a humanoid robot. Information about various people’s identities (appearance, behavior, etc) that we acquire through our daily social experience is so rich and complex that manually encoding them into a database for the robot to memorize is very limiting. In this project, we focus on extending current face technology to allow the robot to opportunistically learn about individuals and their characteristics in an online and unsupervised manner through embodied social interaction. Motivation: The notion of people as distinct individuals plays averyimportant role in our daily social life. If a robot has the ability to recognize and remember people it interacts with, it will be able to learn about characteristics of each individual and treat them uniquely as individuals. This leads to complex social behavior, such as cooperation, dislike, loyalty, and affection. As proposed by [3], if robots have long-term contact with humans, it may be desirable to have them develop individual relationships, which is exactly the aftermath of this social dynamic. Moreover, the ability to distinguish among people allows therobot to build toward more complex social competencies where the idea of people as distinct individuals is crucial, including theory of mind and social referencing.