Cognitive architecture for robot perception and learning based on human-robot interaction

In this paper, we present a cognitive architecture for human-robot interaction to allow robots to perceive, learn and understand the changing state of their surrounding environment. The cognitive architecture is composed of modules for processing of multiple sensory inputs, perception, memory and control of actions. The human-robot interaction is based on the recognition and learning of human faces and actions. Also, a verbal communication between the human and robot is established to control the interaction. Our cognitive architecture has been integrated and tested with the iCub humanoid robot allowing it to recognise and interact with humans, and to understand and learn from their actions.