Conference proceedings article
Qualitative Bayesian Failure Diagnosis for Robot Systems



Publication Details
Authors:
Kirchner, D.; Geihs, K.
Editor:
Burgard, Wolfram
Publisher:
IEEE
Place:
Piscataway, NJ
Publication year:
2014
Pages range:
1-6
Book title:
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), AI and Robotics

Abstract
Reliability is a key challenge for intelligent robotsystems. In order to address this challenge, runtime failuredetection and diagnosis (FDD) is an essential task to maintainautonomous operation. The complexity of fully fledged robotsystems and the included noise in system observations com-plicate this task. In this paper, we present our QualitativeBayesian Failure Diagnosis (QBFD) for precise and robustfailure estimation. Our approach uses a Dynamic BayesianNetwork to model uncertainties of the measurements whileconsidering temporal relations. Instead of detailed a prioriknowledge of system dynamics, our approach models cause-effect relations. These relations are, in practice, more intuitive tospecify. As a consequence, we reduce the level of needed systemknowledge and therefore increase the practical applicability. Weevaluate the quality in respect to two reference approaches inextensive simulations. Due to our results, we are confident thatour proposed approach provides comparable, if not superior,estimation quality, while simultaneously reducing the level ofneeded model details. Furthermore, we provide evidence that,given a proper system decomposition, high quality estimates arepossible using general observations, like the resource usage.

Last updated on 2019-09-01 at 16:21