Object Recognition from Multiple Percepts

Abstract

This paper presents a perceptual system that exploits human caregivers as catalysts for the humanoid robot Cog to perceive and learn about objects, scenes, people, and the robot itself. A broad spectrum of machine learning problems are addressed for object recognition across several categorization levels. The paper introduces a new complex approach to object recognition based on the integration of multiple percepts. Training data for all learning mechanisms is automatically generated from actions by an embodied agent, so that the robot develops categorization autonomously. Cognitive capabilities of the humanoid robot are developmentally created, starting from infant-like abilities for detecting, segmenting, and recognizing percepts over multiple sensing modalities. Human caregivers provide a helping hand for communicating such information to the robot, by acting on the objects, inducing their compliant perception from these human-robot interactions.