Using an adaptive entropy-based threshold for change detection methods – Application to fault-tolerant fusion in collaborative mobile robotics

Abstract

This paper deals with the development of information tools and methods with the main objective of designing a fault-tolerant system, to ensure optimal availability and security when its components no longer fulfill their functions. First, the change detection strategy is reformulated using an entropy-based criterion, allowing the calculation of an adaptive threshold, unlike the Bayes criterion. This approach can be used by any change detection method based on the (generalized) likelihood ratio. In order to validate our approach, we apply the entropy criterion to two commonly used change detection techniques: Cumulative sum (Cusum) and Exponentially Weighted Moving Average (EWMA) control charts. Our strategy is illustrated on a well-known example of the literature. Finally, this entropy-based change detection allows us to design a fault-tolerant fusion methodology, which is experimentally validated from an extended Kalman filter (EKF) in collaborative mobile robotics. Our approach is much more robust than the fixed threshold method with respect to false alarms and missed detections.