Statistical Gas Distribution
Modelling for Mobile Robot
Applications
MATTEO REGGENTE
Computer Science
Örebro Studies in Technology 62 I
ÖREBRO 2014ÖREBRO STUDIES IN TECHNOLOGY 62 2014
MA
tte
o R
eG
Ge
N
te
Sta
tis
tic
al G
as D
is
trib
uti
on M
od
ell
ing f
or M
ob
ile R
ob
ot .
..
Matteo Reggente. I received a Master degree in Engi-neering of Telecommunication from the University of Rome ”La Sapienza”. From 2007 to 2010 I was enrolled as PhD student at Örebro University - Mobile Robotics and Olfac-tion (MRO) Lab. After the PhD studies, I spent two years at the Flemish Institute for Technological Research (VITO). My research interests include mobile robotics, machine lear-ning, environmental monitoring and air pollution. In this dissertation, I present and evaluate algorithms for statistical gas dist-ribution modelling in mobile robot applications.
I derive a representation of the observed gas distribution using geo-refe-renced gas concentration measurements collected with mobile robots. On the robots, I mounted different sensors. I used laser scanner for the autonomous robot navigation, localization and obstacle avoidance; gas sensors for gas measurements, and ultrasonic anemometer for wind measurements. The proposed algorithms fuse the readings from all the sensors to build the gas distribution model in the environment using a localized Gaussian kernel. The kernel adapts is shape and orientation according to the local wind measure-ments. The algorithms provide four 2D or 3D grid maps. The weight map is a graphical representation of the density of measurements; the confidence map, highlights areas in which the model is considered being trustful; the map of the mean gas distribution is a graphical representation of the modelled gas distribution in the monitored environment; the map of the variance estimate gives a graphical representation of the spatial structure of the variance of the mean estimate.
Ground truth evaluation is a critical issue for gas distribution modelling with mobile platforms because it is not possible at the scale and for the che-micals that we are interested in “to take a snapshot” of the instantaneous concentration field. To cope with this problem, I proposed two methods to evaluate and compare gas distribution models. Firstly, I created a ground-truth gas distribution using an original simulation environment, and I compared the models with this ground truth gas distribution. Secondly, considering that a good model should explain the measurements and accurately predict future measurements, in the real world applications, I evaluated the models according to their ability in inferring unseen gas concentrations.
issn 1650-8580 isbn 978-91-7529-034-8
