Valencia, 29 de Septiembre de 2009.


Programa de Doctorado:
Automática e Informática Industrial
Autor de la tesis:
D. Danilo A. Navarro García
Titulo de la tesis:
Contribución a la autolocalización de robots móviles basada en la 
fusión de información multisensorial
Director de la tesis:
Dr. Gines Benet Gilabert

Abstract

 
	At past, robots were working in environments specially prepared for them. Every 
component of their work space was placed in a predefined position and orientation, in order to 
robots knew with precision the place where they were. At present many applications require that 
robots have autonomy features as for example: skill to identify by means of their sensors the 
environment characteristics, capability for self localization, and aptitude to move along of their 
work space, paying attention to the temporary variations arising from the environment. In spite of 
the results obtained in the robots localization and mapping area are abundant and very significant; 
even so exist problems for solving those that are specially linked to the use of low-cost sensors. 
This is because the data obtained with these kind of sensors is noisy and very uncertain. Aimed to 
contribute towards to solve the problem mentioned before, this thesis focus on the modelling of 
low-cost sensors which are typically used in mobile robotics (Optical encoder, ultrasounds and 
infrared sensors, magnetic compass), thus as on the studying of how they could be used in auto-
location and map building problems. 

By mean of information fusing, a mobile robot using low-cost sensors is capable of estimate its 
position adequately, and in this way navigate reliably in structured environments. In order to test 
off-line the different mechanisms of merging and filtering that we propose, along this thesis we 
develop the sensory models of three low-cost sensors which are frequently used in mobile robotics: 
ultrasonic and infrared range finder, and odometric dead-reckoning sensor. Furthermore we develop 
the pseudo code that is necessary for the embedded operation of each one of these sensors on a 
generic real robot. In addition we formulate the procedures and synthesize the software to be 
embedded in the real robot YAIR(Yet Another Intelligent Robot) for purpose of navigation control, 
acquisition and filtering of sensor data, and for merging sensory information related to environment 
mapping and robot localization. The models on this way developed allow incorporate they to 
another advanced systems intended for density map or featured map construction. Likewise, from 
the models mentioned before we have developed two localization systems: the first one uses 
odometric information for Dead-Reckoning, whereas the second one uses walls and corner as 
natural Landmarks and an Extended Kalman Filter approach for correction of the robot position.

In order to evaluate and validate the proposed methods, we carried out mapping and localization 
experimental test using a robot called YAIR which is a multi-sensorial differential-drive robot. 
Because this robot is a real time system operating in a real environment, the analysis and validation 
of the results was based on the direct observation of the robot behaviour when it was exploring the 
environment or navigating on it. The results showed that with the information merging approach the 
quality of the environment maps obtained was improved significantly, likewise the precision of the 
robot localization process. Moreover, with the localization system developed in this thesis, the 
localization error was reduced from 3 meters to few centimetres in distances about 80 meters long.