Integrating Multiple Sources of Knowledge for the Intelligent Detection of Anomalous Sensory Data in a Mobile Robot

Abstract

For service robots to expand in everyday scenarios they must be able to identify and manage abnormal situations intelligently. In this paper we work at a basic sensor level, by dealing with raw data produced by diverse devices subjected to some negative circumstances such as adverse environmental conditions or difficult to perceive objects. We have implemented a probabilistic Bayesian inference process for deducing whether the sensors are working nominally or not, which abnormal situation occurs, and even to correct their data. Our inference system works by integrating in a rigorous and homogeneous mathematical framework multiple sources and modalities of knowledge: human expert, external information systems, application-specific and temporal. The results on a real service robot navigating in a structured mixed indoor-outdoor environment demonstrate good detection capabilities and set a promising basis for improving robustness and safety in many common service tasks.