Autonomy in Cognitive Development of Robots: Embracing Emergent and Predefined Knowledge and Behavior

Abstract

This article addresses the fundamental questions on machine learning: what does it mean for machines to learn from experience, and what does it mean by machines in machine learning? Despite recent popularity and growth, significant challenges remain as the industry rapidly advances toward autonomous machines. In the context of autonomy, there is more to learning from experience than training machines to approximate big data. This brings the fundamental questions to the forefront of automation science and engineering as a critical area of exploration. This article examines the precise notion of autonomy in the context of machine learning and provides a general framework for cyber-physical systems to become fully autonomous by learning from experience. The framework is derived from the principles of developmental autonomous behavior, which encapsulates broad classes of learning mechanisms. It offers a novel use case of machine learning where sensorimotor systems build inference engines internally on their own by their own initiatives to develop new skills and behavior. The key contributions of this article are threefold. First on knowledge: it provides precise definitions of emergent and predefined knowledge and their roles in cognitive development of machines. Second on autonomy: it clarifies what a fully autonomous machine means by providing the precise definitions of autonomy and emergent behavior. Third on machine learning: it unifies machine learning as the ultimate-proximate causal drivers of emergent behavior. Ultimately, this article logically explains why and how a fully autonomous machine is possible by directly answering the fundamental questions.