An Open and Modular Architecture for Autonomous and Intelligent Systems

Abstract

Over the past few decades, remarkable progress has been made in the the field of Artificial Intelligence (AI). For some tasks such as games, natural language processing, and image classification, AI powered applications can match or surpass human performance. Despite this progress, truly Autonomous and Intelligent Systems (AIS) serving the needs of, and sharing the environment with humans, are yet to become a commercial reality. AIS engineering requires considerable integration efforts that must be disciplined and guided by a reference model enabling reuse and concurrent design: an open and modular architecture. While several specialized architectures have been developed over the course of few decades, there is a need for a unified approach that supports multi-agency, learning, knowledge representation, reasoning, planning, and run-time verification. In this paper, we propose an open and modular architecture for autonomous and intelligent systems. We start by defining the three primary modules of the architecture, namely situational assessment, knowledge repository and management, and decision making. We then refine each module into functional units and we describe possible interaction patterns among them.