COBRA - A generic architecture for robust treatment of uncertain information

Abstract

This paper introduces the COBRA-architecture as an approach to handle uncertainties in autonomous robots and embedded systems in order to increase their robustness. In the COBRA-architecture, the trustworthiness of information is there for modeled explicitly by meta-signals, called trust signals. At a coarse-grained level, the architecture also extends normal functional modules to a triplet of modules, one containing the normal functionality with handling of deficient information, one containing a fall-back strategy for the case of too insufficient information, and one mitigating between them depending on the respective trustworthiness. The COBRA-architecture incorporates the trust signals also at a fine-grained level into the basic signal processing. As an example the extension of a typical class of continuous functions is presented to achieve more trusty results and to generate also a trust signal for the output. Both basic aspects of the COBRA-architecture are illustrated by the control architecture of the walking robot WALTER. © 2013 Lecture Notes in Informatics (LNI), Proceedings - Series of the Gesellschaft fur Informatik (GI).