Self-awareness in real-time cognitive control architectures
Ricardo Sanz, Ignacio López Paniagua, Carlos Hernández · National Conference on Artificial Intelligence · 2007
In this paper we describe the approach taken in the UPM Autonomous Systems Laboratory for the development of technology of full, bounded autonomy. This effort is taking us into the study of the phenomenology of consciousness especially in relation with the construction of mechanisms for system self-awareness. The progress of this work can be tracked in our ASys and SOUL projects webpages. The ASys Project is a long-term project that deals with the development of an engineering process and a collection of assets for the implementation of autonomous systems. The SOUL Project develops a generic architecture for self-aware autonomous systems. The Frontiers of Control Technology The field of control technology addresses the different engineering aspects involved in the construction of control systems. Control systems are subsystems designed to improve operation of certain systems of interest constituting an integral part of them. There are plenty of control systems out there even when in most cases they may pass unnoticed to the point that Karl Astrm, a major control guru, said of this technology that it is a hidden technology. Control systems can be found in the heating of our homes, in the landing gear for planes, in our CD players, in heart pacemakers or insulin pumps, in keeping the electrical network stable, etc. In a sense, control systems do keep working in sufficiently good condition the technological infrastructure of our modern lives. Research in control systems (see for example www.ifaccontrol.org) has been dealing with the different aspects of this technology and the science that supports it. From the theoretical underpinnings of control systems to the purely practical applications in several domains the field of control has always dealt with the issue of autonomy. In many cases limited to the system being able to address certain limited classes of perturbations, but there are many reasons for pursuing the objective of fully autonomous machines. ∗We acknowledge the support of the Spanish Ministry of Education and Science through grant C3: Control Consciente Cognitivo and the European Commission thorough Grant ICEA: Integrating Cognition, Emotion and Autonomy. Copyright c © 2007, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. Reducing operating costs and improving system performance —that is maximizing the benefit— were the main factors behind the drive for improved automation in the past. During the last years, however, a new force is gaining momentum: the need for augmented dependability of complex systems. This is a problem that classical control theory nor standard software engineering cannot deal with. This is the reason utimately grounding the very core of our research into the deeps of artificial consciousness. Autonomy for cost reduction and performance In many cases, automated system performance is much higher than manual system performance. This is due to many reasons: 1) Automated systems do not get tired; 2) automated systems are usually faster; 3) automated systems are more precise; etc. There are even technical systems that cannot be manually operated at all due to the intrinsic limitations of humans — for example in high-speed machining— or practical or legal issues typically associated with worker health—for example in space exploration, foundries, chemical plants, etc. In many practical cases the problem of building a controller for a well known production process in a well controlled environment is mostly solved; only minor details persist related with continuous optimisation. The problem for control engineering appears with uncertainty. When the plant is not so well known, or when the operational and/or environmental conditions are of high uncertainty the traditional control system strategy fails. This is so because having a good knowledge of the plant is the first necessary step to building a good controller for it (Conant & Ashby 1970). There are many textbooks on controller design in general terms, centered in particular kinds of controller designs or centered in concrete application domains. In the last case, the domain typically constrains the kind of implementation that a controller may have (e.g. table-driven controllers in resource constrained electronic control units in vehicles or software-less controllers in some safety-critical applications). Industrial-level control system engineering is a well established professional practice that addresses the task of controller construction in six steps: