A Cognitive Architecture for Social Robots

Abstract

The paper illustrates a software architecture allowing a robot to socially interact with human beings, sharing with them some basilar cognitive mechanisms. Robust sensing of the environment and people is strongly linked with an artificial somatosensory system that drives the robot behavior at a low level and influences its motivation. Both long-term memory and short-term memory store relevant data to detect and recognize the social context (and social practice), and the human social behavior. Using both internal and external evaluations, the robot learns and improves its social skills, which take into account its physiological and emotional demands (affiliation, competence, certainty). Social interaction is encoded in the cognitive architecture by considering at the same level the human understanding and the robot communicative actions. This is done by using the same interaction channels (both verbal and nonverbal). Some examples derived from previous works show the effectiveness and the potential of the cognitive architecture.