Improved Resilience Controllers Using Cognitive Patterns

Abstract

Construction of robotic controllers has been usually done by the instantiation of specific architectural designs. The ASys design strategy described in this work addresses the synthesis of custom robot architectures by means of a requirements-driven application of universal design patterns. In this paper we present three of these patterns —the Epistemic Control Loop, the MetaControl and the Deep Model Reflection patterns— that constitute the core of a pattern language for a new class of adaptive and robust control architectures for autonomous robots. A reference architecture for self-aware autonomous systems is synthesised from these patterns and demonstrated in the control of an autonomous mobile robot. The term “autonomous” gains a deeper significance in this context of reflective, pattern based controllers.