Machine Autonomy and Developmental Autonomous Behavior
The core objective of this research is to derive and apply a principled and practical framework to understand developmental autonomous behavior (DAB) of machines. DAB refers to “the general ability of a machine to acquire new skills and behavior from its birth to maturity on its own without human intervention” [1]. It draws inspiration from ethology, the study of animal behavior, to derive the ultimate-proximate explanations of the causality, mechanisms, and circumstances of behavior emergence that are grounded in physical interactions and sensory experiences of machines in their environment.
The key difference of this research from the related fields, such as cognitive science, artificial intelligence, and developmental robotics, is its objective and perspective. Unlike others, this research does not aim to understand human cognition or build a machine that replicates human capabilities. Instead, it strictly focuses on identifying the systems and processes that achieve autonomy from the perspective of machines. Because of this distinction, this research is free from the obligation to justify theories in terms of intangible, ill-defined anthropocentric and subjective concepts, such as beliefs, cognition, consciousness, intelligence, and sentience. By leaving those concepts out of the scope, this research is free to close the loop between deductive and inductive analyses of machine behavior with direct access to the machines’ internals, which is a major obstacle in human studies. This research deepens our understanding of autonomous machines, which is paramount to the safety of our society and thus a critical area of exploration for the advancement of automation science and engineering.
The concept of autonomy is central in this research. Based on the principles of embodiment and situatedness, the proposed framework defines autonomy in the ecological sense of survival and operation continuity. Machine autonomy is therefore tested in terms of the ability to maintain its operation in perpetuity, including the ability to find a way to supply its own energy without human intervention. The framework posits that value systems for the purpose of autonomy drive successive development of memory functions, resulting in progressive changes in behavior from innate reflexive to episodic, procedural, and autonomic behavior. The framework also derives a logical explanation for the transformation process that enables a physical sensorimotor system to become a symbol-like concept processor, fostering conceptual and social behavior development. By precisely defining the concept of autonomy and the role of value systems, the framework provides the principles of developmental autonomous behavior emerging from experience and learning.
The framework is structured for utility to build and analyze machines that exhibit developmental autonomous behavior by incorporating two practical tools: machine learning (ML) and cyber-physical systems (CPS). It broadens the role of ML algorithms from data approximators to self-learning systems by exploiting the algorithmic attributes, thereby contributing to various aspects of behavior emergence in autonomous machines. It also provides a general architecture that leverages distributed computing and communication capabilities of CPS, necessary for continuous autonomous operation in practical applications.
A thread of research begins with the fundamental principles described in [1]. The framework was demonstrated in a minimally configured mobile robot in [2]. The key concepts on taxonomy of autonomy and ML algorithms for behavior emergence were described in [3]. A general taxonomic classification and identification system for robots and autonomous systems was proposed in [4][5]. Current topics include reference-based machine learning for conceptual behavior emergence and structural analyses of autonomous systems [6].
[1] S. Isaka, “Developmental Autonomous Behavior: An Ethological Perspective to Understanding Machines,” in IEEE Access, vol. 11, pp. 17375-17423, 2023.
[2] S. Isaka, “An Ethological Analysis of Developmental Behavior in Machines," in Proc. 2023 IEEE Int. Conf. on Development and Learning (ICDL), Macau, China, pp. 79-86, 2023.
[3] S. Isaka, “Autonomy in Cognitive Development of Robots: Embracing Emergent and Predefined Knowledge and Behavior,” to appear in Proc. 2024 IEEE 20th Int. Conf. on Automation Science and Engineering (CASE), Bari, Italy, 2024.
[4] S. Isaka, “A Taxonomic Classification and Identification System for Robots: Abstract,” to be presented at the 2024 IEEE Int. Conf. on Systems, Man, and Cybernetics (SMC), Kuching, Malaysia, 2024.
[5] S. Isaka, “A Taxonomic Classification and Identification System for Robots and Autonomous Systems,” Submitted for publication.
[6] S. Isaka, “Behavior-Based Robotics, Cyber-Physical Systems, and Machine Learning: United as a Practical Pathway Toward Machine Autonomy,” unpublished.