Robot controllers for highly dynamic environments with real-time constraints [Elektronische Ressource] / vorgelegt von Alexander Antoine Ferrein

Sujets

• Informatik

Informations

• Publié par: rheinisch-westfalischen_technischen_hochschule_-rwth-_aachen
• Publié le: 01 janvier 2007
• Nombre de lectures: 29
• Langue: English
• Poids de l'ouvrage: 2 Mo

Extrait

Robot Controllers for Highly Dynamic Environments
With Real-time Constraints
¨ ¨Von der Fakultat fur Mathematik, Informatik und Naturwissenschaften der
Rheinisch-Westfa¨lischen Technischen Hochschule Aachen zur Erlangung des
akademischen Grades eines Doktors der Naturwissenschaften genehmigte Dissertation
vorgelegt von
Diplom-Informatiker
Alexander Antoine Ferrein
aus Hilden
Berichter: Universita¨tsprofessor Gerhard Lakemeyer, Ph.D.
Universita¨tsprofessor Michael Beetz, Ph.D.
Tag der mu¨ndlichen Pru¨fung: 20. Dezember 2007
Diese Dissertation ist auf den Internetseiten der Hochschulbibliothek verfu¨gbar.TO MY FAMILYAcknowledgements
First of all, I would like to thank my supervisor Gerhard Lakemeyer for several reasons. He
convinced me to start the quest on mobile robots and with this giving me this great opportunity and
experience of being part of the RoboCup project. During the recent years he always gave the right
advice at the right time in his very subtle way. His way of giving sublte (mostly) reinforcement
(positive as well as negative) very much enforced my scientific education and development, and
I would like to thank him for the trust that our efforts will come to some good end, and for his
patience and understanding that sometimes making things run in reality takes longer than expected.
In this context I have to underline that I am very grateful for his respect of the practical work we
have done during the last years. His passion to sit in early morning hours with the RoboCup team
to debug something at some tournament site is well remembered. Finally, I would like to thank
him for his valuable comments on earlier versions of this thesis, helping to make it a coherent and
hopefully (formally) sound document. I am grateful to Michael Beetz who kindly accepted the
task of the second reviewer of this thesis.
I’m indebted to many people supporting this work during the last years. It may sound a little
bit like an acceptance speech at the Grammys, but I do owe a lot to the following people. I
have to thank the Knowledge-based Systems Group for the great support of our endevours, in
particular (in the order of appearance) Gero Iwan, Hendrik Grosskreutz, Gunte¨ r Gans, Frank Dylla,
Christian Fritz, Vazha Amiranashvili, Jens Classen, Stefan Schiffer, and Daniel Beck; Dominik
Schmitz for listening to my complaints on many train rides, for his valuable comments and solution
ideas on several problems; and the whole group of people working at Computer Science V. I
have to thank all the members of the AllemaniACs RoboCup team. The members over the last
five years were (in alphabetical order): Daniel Beck, Florin Bley, Masrur Doostdar, Frank Dylla,
Benjamin von Eicken, Christian Fritz, Carsten Gester, Claudia Gonne¨ r, Lutz Hermanns, Yuxio Hu,
Stefan Jacobs, Mirko Kohns, Savas Konur, Alexandros Matsikis, Christoph Mies, Tim Niemulle¨ r,
Martin Rous, Stefan Schiffer, Nils Springob, Andreas Strack, Pei Tang, Philipp Vorst, and Jost
Wunderlich. With such a long list of people one cannot pinpoint the reasons for my gratefulness
in detail. I just want to remark that some of these people I owe much and that these people
should know that I know! Thanks for the good times, but also for the hard times we went through
together. I was able to learn a lot from all of them in so many respects; the working together
(which was not always a pleasure) let me learn so much and will be well remembered. I would
also like to thank the AllemaniACs@Home team (Stefan and Tim and Masrur) explicitly for their
outstanding performance during last RoboCups showing that our robots are not only bulky but that
the approach works in practice!
Next, I want to thank the German Science Foundation (DFG) and the Ministry of Science and
Education (MSWF) of North-Rhine Wesfalia for the financial support of the RoboCup project.
Further, I’m also grateful for the financial support of Computer Science V, personified through
Mathias Jarke and Gerhard Lakemeyer, and the Bonn-Aachen Institute of Technology.
Thanks to Tom Waits and to Linkin Park for providing the right background atmosphere during
the long hours of the writing this document.
I am deeply indebted to Tessa Oliver inter alia for proof-reading the whole thesis. Her com-
ments and advice finally made this document ’human-readable’. I want to thank my family for
their support during recent years. In the end, I have to thank Tanja for her love and her sup-
port especially during the time of writing this thesis, always having understanding for “I’d like to
but, . . .”.Abstract
The fields of mobile autonomous robotics and cognitive robotics are active research fields. In
the last decade progress has become more and more visible. The reason for this is that the sys-
tems became more affordable, there were side effects to automobile industries (driving assistance
systems) which make use of similar methods, and that activities like RoboCup (robotic soccer)
became more present in the media. Additionally, there are efforts to standardize software and
hardware components. In recent years several successful applications showed that mobile robots
can interact with their environment and fulfill meaningful and useful tasks. Nevertheless, many
questions on how to design autonomous mobile robots remain open and are subject to active re-
search in this field. For problems like navigation, collision avoidance, and localization, robust
approaches have been proposed which are widely used. The question of how such robots can act
intelligently, has wide-spread ideas and approaches.
In this thesis we propose an approach to the problem of intelligent decision making (delibera-
tion) for robots or agents which moreover have to decide under real-time constraints in adversarial
domains, i.e. multi-agent domains where opponents have contrary goals and try to foil the goals
of the opposing team. Agents can be seen as non-embodied robots acting in simulated environ-
ments. Our account is based on the logical agent high-level programming language Golog. Golog
is based on the situation calculus, a powerful logical calculus for reasoning about actions and
change. During recent years several extensions for integrating concurrency and sensing, dealing
with continuous change, and applying decision theory have been made. The question we are con-
cerned with in this thesis is how these different approaches can be put together in such a way that
an agent is enabled to come to intelligent decisions while acting in dynamic, adversarial real-time
domains. The crucial point is that deliberation in general is computationally expensive, but nev-
ertheless is needed to come to intelligent decisions. On the other hand, the robot must be able to
react quickly to changes in the environment, because otherwise an opponent might take advantage.
We propose the language READYLOG as an approach to intelligent decision making especially
in dynamic real-time domains. READYLOG is a GOLOG family language and combines features
known from other dialects in one framework. It is based on the situation calculus and offers proba-
bilistically projecting a given world situation into the future, dealing with a continuously changing
world, coming with an efficient implementation and a mechanism to progress the internal database.
READYLOG especially makes use of decision-theoretic planning as a means to intelligent decision
making. Several alternatives in the robot program are left open and R EADYLOG chooses the most
promising one against a background optimization theory. As decision-theoretic planning is com-
putationally costly, we are looking for ways to reduce the complexity of planning. One possibility
which we propose is to make use of options, which are macro-actions in the decision-theoretic
context.
We show several detailed applications of READYLOG in dynamic real-time domains. These
range from interactive computer games to robotic soccer, simulated as well as with real robots.
READYLOG has been successfully applied for several soccer agents which participated in the in-
ternational RoboCup world championships. As we are also dealing with real robots, we also
discuss several new approaches and extensions in the field of robotics, like a new robust laser-
based navigation algorithm. Finally, as READYLOG comes with a formal logical semantics it is
well-suited to model behavior in a general way. We started to develop a formal soccer theory for
robots based on human soccer knowledge. The formalization language we used is READYLOG.Zusammenfassung
Die Forschungsfelder Mobile Autonome Robotik“ und Kognitive Robotik“ sind aktive Fel-
” ”
der und werden intensiv bearbeitet. Der Fortschritt auf diesen Gebieten ist besonders in den letzten
10 Jahren sichtbar geworden. Ein Grund fur¨ die versta¨rkte Wahrnehmbarkeit dieser Forschungs-
gebiete ist, dass Robotersysteme guns¨ tiger in der Anschaffung und Entwicklung werden. Es
gibt vermehrt Querbeziehungen zur Automobilindustrie im Bereich von Fahrerassistenzsystemen,
wo a¨hnliche Methoden fur¨ a¨hnliche Problemstellungen eingesetzt werden. Davon profitiert die
Roboterforschung u.a., da durch die Massenproduktion in der Automobilindustrie z.B. Sensorik
billiger wird. Weiterhin sind Initiativen wie RoboCup, die Weltmeisterschaft der Fußballroboter,
sta¨rker pra¨sent in den Medien. Darube¨ r hinau