Business-management-inspired sensor data fusion

Abstract

We apply a new type of algorithm for sensor data fusion that was originally developed for estimation of business indicators. The origin of the MCMC algorithm SamPro is the consideration of uncertainty in business indicators, such as profit, sales, and cost, which results from measurement errors or forecasting. Furthermore, the SamPro algorithm uses model-based redundancy to generate virtual measurements; it is able to cope with and can reduce uncertainty of metrical data, including different and even nonparametric data distributions. In this paper, we present an adaptation of the algorithm focused on (distributed) sensor measurements. In such scenarios, the information redundancy bases on multi-modal sensors. Those results can be fused directly or after model based transformations. We validate our approach in a localization scenario fusing laser distance measurements, camera images, and on-board odometry to estimate the current position of a mobile robot. For this purpose we utilize sensor models for each sensor, including specific sensor faults and noise behavior, to generate and fuse virtual sensor measurement.