Contributions to event-based state estimation for intelligent spaces

  1. MARTÍNEZ REY, MIGUEL
Supervised by:
  1. Alfredo Gardel Vicente Director
  2. Felipe Espinosa Zapata Director

Defence university: Universidad de Alcalá

Fecha de defensa: 12 June 2017

Committee:
  1. Enrique Santiso Gómez Chair
  2. Roberto Iglesias Rodríguez Secretary
  3. Héctor García de Marina Committee member

Type: Thesis

Teseo: 531570 DIALNET lock_openTESEO editor

Abstract

An intelligent space is an environment equipped with sensors that provide services to the people or machines that occupy it, such as identification, localisation, voice or gestures recognition, etc. Elements of an intelligent space are often connected by wireless technologies, and are battery powered. In this context, it is important to reduce the use of sensors in order to extend the battery life and preserve the communication bandwidth. In order to achieve this goal, one alternative is to acquire and send information only at the time instants when they are really necessary. Opposed to traditional time-based systems, that are driven by a clock signal, this new paradigm is referred to as event-based estimation, which is receiving a growing attention in research in the last years. This work is the combination of three contributions in the field of event-based state estimation for sensor networks. Each one of them corresponds to one of the main elements of a state estimation system, which are the sensors, the estimator and the network communication. Concerning the sensors, two new sampling methods based on the Mahalanobis distance were proposed. The idea is to evaluate the importance of a sample considering the distance of the performed correction in the case that the sample is applied by the estimator. For this purpose, the Mahalanobis distance is a metric that takes into account the estimation error probability distribution as well as the measurement noise. The proposed methods are compared with known sampling alternatives in simulation and also in experimental tests, where it was applied to an indoor localisation task. Concerning the estimator, an adaptive variance-based sampling algorithm is applied to assist the guidance of a robotic unit. In the proposed method, the estimator module requests samples to the sensors whenever the position and orientation uncertainties reach a certain value, so that the expected estimation error is bounded. Additionally, the threshold value adapts to the controller needs, requiring less precision when the distance towards the desired position is large. The proposal is validated in simulation and with experimental tests on a mobile robot, where a considerable reduction of communication was achieved without a noticeable degradation of the guidance manoeuvre. Concerning the communication network, the effect of delayed messages on estimation performance was analysed, considering that such delays are in turn caused by occupation produced by sensor transmissions. We were able to obtain the optimal sampling rate for the ideal case where the whole state vector can be sensed without noise. This result is extrapolated to the general case where, under certain conditions, an upper bound of the solution can be guaranteed.