Sensor fusion methods for indoor navigation using UWB radio aided INS/DR

Abstract: Some applications such as industrial automation, cargo handling, warehouse managing,monitoring of autonomous robot or personnel localization, require reliable indoor positioning. In particular, the requirement is to accurately localize a mobile wireless node in real-time. In outdoor scenarios, Global Navigation Satellite Systems (GNSSs) are commonly used for positioning. Nevertheless, they present notable shortcomings for being used in indoor applications. The principal disadvantages are the low accuracy achieved and the attenuation and reflections introduced by buildings and other structures, which make impossible their use in most cases. The fixed and smaller operational area of an indoor positioning application in comparison with the area of an outdoor application makes it possible to install a radio positioning infrastructure. Concretely, the Ultra Wide Band (UWB) positioning system considered in this thesis consists of nodes installed in fixed and known positions (slaves) and a mobile node of unknown position (master) which is the one to be localized. The operation of the system is based on calculating the distance from the master to the slaves by measuring the Round-Trip-Time of a UWB pulse. In this way an estimate of the position can be obtained, with the advantage of having bounded errors,but, in contrast, the estimations have a low dynamic range and other navigation information cannot be measured. The master has some sensors attached, an Inertial Measurement Unit (IMU) and a Dead-Reckoning (DR) system. By propagating the navigation equations, estimates of the position and other navigational states can be obtained, but with the disadvantage of having errors that grow with time. The solution of the problems related to the individual location systems is using information fusion algorithms. Such algorithms, by means of integrating different sources of data, in this case the UWB range measurements and the IMU and DR measurements, create a positioning system which combines the benefits of the individual systems. In this project the aim is to develop, implement and evaluate sensor fusion algorithms. In particular, the developed algorithms are based on the Extended Kalman Filter and the Particle Filter. Furthermore, they have been adapted to different dynamic models, in order to find the algorithms which fit better with the motion of a mobile node. The developed algorithms have been tested with simulated trajectories using Matlab, and with real experimental datasets acquired by a mobile robot. The results show the benefits of the information fusion, since the accuracies obtained in the estimations outperform the accuracies obtained with the individual systems. Also in the most unfavorable cases, when one of the sources has high errors in the measurements, the algorithms are able to discard the useless information and estimate using only the useful measurements, proving the robustness of the system.