{"id":{"repo_id":"freiburg-diss","oai_identifier":"oai:freidok.uni-freiburg.de:1362"},"canonical_url":"https://search.dev.ndltd.org/etd/freiburg-diss/oai:freidok.uni-freiburg.de:1362","repository":{"repo_id":"freiburg-diss","name":"University of Freiburg","base_url":"https://freidok.uni-freiburg.de/oai/oai2.php"},"display":{"title":"Mobile robot navigation in dynamic environments","abstract":"Service robots which act in environments populated by humans have
become very popular in the last few years. A variety of systems exists
which act for example in hospitals, office buildings, department
stores, and museums. Furthermore, several multi-robot systems have
been developed for tasks which can be accomplished more efficiently by
a whole team of robots than just by a single robot. These tasks
include surface cleaning, deliveries, and the exploration of unknown
terrain. Whenever teams of mobile robots are operating in the same
environment their motions have to be coordinated in order to avoid
congestions or collisions. At the same time the robots should perform
their navigation tasks in a minimum amount of time. Thus,
sophisticated path planning techniques are needed that fulfill these
requirements. Since the joint configuration space of the robots is
typically huge and grows exponentially with the number of robots,
existing path planning methods for single robot systems cannot
directly be transferred to multi-robot systems.
Many existing path planning methods for multi-robot systems are
decoupled, which means that they first plan paths for the individual
robots independently. Afterward, they check if the robots would get
too close to each other if the paths were executed. In such a case the
paths are recomputed to avoid these conflicts. Many decoupled methods
assign priorities to the individual robots. These priorities define an
order in which the paths of the robots have to be recomputed. By
computing the path of a robot, the paths of the robots with higher
priority are considered as fixed. This way, the size of the search
space is extremely reduced. Most of the existing prioritized decoupled
methods use a fixed priority scheme (order of the robots). However,
the order in which the paths of the robots are recomputed has a
serious influence on whether a solution can be found at all and on how
efficient the solution is for the overall multi-robot system.
In the first part of this thesis we present an approach which searches
in the space of all priority schemes to find an order of the robots
for which a solution to the path planning problem can be computed.
During the search, we utilize constraints between the priorities of
the robots which are automatically derived from the task
specification. After an appropriate priority scheme has been found,
our technique tries to improve it by using a hill-climbing strategy.
Our search method can be used to find and optimize paths generated by
any prioritized path-planning technique. In several experiments with a
real-robot system as well as in simulation we show that our approach
produces efficient solutions even for difficult path planning
problems.
The second part of this thesis is focused on robots acting in
environments populated by humans. These systems can improve their
behavior if they react appropriately to the activities of the
surrounding people and do not interfere with them. In contrast to a
multi-robot path planning system, the future movements of people are
not known. Therefore, the robots have to be able to detect people
with their sensors, to identify them, and to learn their intentions in
order to be able to make better predictions of their future behavior.
In this thesis we present an approach to learn typical motion patterns
of people from sensor data using the EM algorithm. Furthermore, we
describe how the learned patterns can be used to predict future
movements of the people. Afterward, we explain how this knowledge can
be integrated into the path planning process of a mobile robot.
Finally, we introduce a method which automatically derives Hidden
Markov Models (HMMs) from the learned motion models. These HMMs can be
used by a mobile robot to predict the positions of multiple persons
even when they are outside its field of view. To update the HMMs based
on laser-range data and vision information we apply Joint
Probabilistic Data Association Filters. In practice, the robot becomes
uncertain about the positions of people if it does not observe them
for a long period of time. We therefore propose a decision-theoretic
approach to determine observation actions that are carried out while
the robot is executing its tasks. Practical experiments carried out
with our mobile robot demonstrate that our method is able to learn
typical motion patterns of people, that the navigation behavior of the
robot can be improved by predicting the motions of people based on the
learned motion patterns, that the derived HMMs can be used to reliably
maintain a probabilistic belief about the current positions of
multiple persons even if they are currently not in its field of view,
and that our technique generates effective actions that seriously
reduce the uncertainty in the belief about the positions of people.","abstract_html":"Service robots which act in environments populated by humans have <br>become very popular in the last few years. A variety of systems exists <br>which act for example in hospitals, office buildings, department <br>stores, and museums. Furthermore, several multi-robot systems have <br>been developed for tasks which can be accomplished more efficiently by <br>a whole team of robots than just by a single robot. These tasks <br>include surface cleaning, deliveries, and the exploration of unknown <br>terrain. Whenever teams of mobile robots are operating in the same <br>environment their motions have to be coordinated in order to avoid <br>congestions or collisions. At the same time the robots should perform <br>their navigation tasks in a minimum amount of time. Thus, <br>sophisticated path planning techniques are needed that fulfill these <br>requirements. Since the joint configuration space of the robots is <br>typically huge and grows exponentially with the number of robots, <br>existing path planning methods for single robot systems cannot <br>directly be transferred to multi-robot systems. <br> <br>Many existing path planning methods for multi-robot systems are <br>decoupled, which means that they first plan paths for the individual <br>robots independently. Afterward, they check if the robots would get <br>too close to each other if the paths were executed. In such a case the <br>paths are recomputed to avoid these conflicts. Many decoupled methods <br>assign priorities to the individual robots. These priorities define an <br>order in which the paths of the robots have to be recomputed. By <br>computing the path of a robot, the paths of the robots with higher <br>priority are considered as fixed. This way, the size of the search <br>space is extremely reduced. Most of the existing prioritized decoupled <br>methods use a fixed priority scheme (order of the robots). However, <br>the order in which the paths of the robots are recomputed has a <br>serious influence on whether a solution can be found at all and on how <br>efficient the solution is for the overall multi-robot system. <br> <br>In the first part of this thesis we present an approach which searches <br>in the space of all priority schemes to find an order of the robots <br>for which a solution to the path planning problem can be computed. <br>During the search, we utilize constraints between the priorities of <br>the robots which are automatically derived from the task <br>specification. After an appropriate priority scheme has been found, <br>our technique tries to improve it by using a hill-climbing strategy. <br>Our search method can be used to find and optimize paths generated by <br>any prioritized path-planning technique. In several experiments with a <br>real-robot system as well as in simulation we show that our approach <br>produces efficient solutions even for difficult path planning <br>problems. <br> <br>The second part of this thesis is focused on robots acting in <br>environments populated by humans. These systems can improve their <br>behavior if they react appropriately to the activities of the <br>surrounding people and do not interfere with them. In contrast to a <br>multi-robot path planning system, the future movements of people are <br>not known. Therefore, the robots have to be able to detect people <br>with their sensors, to identify them, and to learn their intentions in <br>order to be able to make better predictions of their future behavior. <br>In this thesis we present an approach to learn typical motion patterns <br>of people from sensor data using the EM algorithm. Furthermore, we <br>describe how the learned patterns can be used to predict future <br>movements of the people. Afterward, we explain how this knowledge can <br>be integrated into the path planning process of a mobile robot. <br>Finally, we introduce a method which automatically derives Hidden <br>Markov Models (HMMs) from the learned motion models. These HMMs can be <br>used by a mobile robot to predict the positions of multiple persons <br>even when they are outside its field of view. To update the HMMs based <br>on laser-range data and vision information we apply Joint <br>Probabilistic Data Association Filters. In practice, the robot becomes <br>uncertain about the positions of people if it does not observe them <br>for a long period of time. We therefore propose a decision-theoretic <br>approach to determine observation actions that are carried out while <br>the robot is executing its tasks. Practical experiments carried out <br>with our mobile robot demonstrate that our method is able to learn <br>typical motion patterns of people, that the navigation behavior of the <br>robot can be improved by predicting the motions of people based on the <br>learned motion patterns, that the derived HMMs can be used to reliably <br>maintain a probabilistic belief about the current positions of <br>multiple persons even if they are currently not in its field of view, <br>and that our technique generates effective actions that seriously <br>reduce the uncertainty in the belief about the positions of people.","abstract_has_math":false,"creators":["Bennewitz, Maren"],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":["Burgard, Wolfram"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":null,"date_issued":"","date_published":null,"updated_at":"2026-07-24T02:22:10Z","subjects":["Mehrroboternavigation","Lernen von Bewegungsmustern von Personen","Roboternavigation","Zustandsschaetzung","Vorhersagen von Personenbewegungen","Multi-robot navigation","Learning motion patterns of people","robot navigation","state estimation","predicting the motions of people"],"languages":[],"rights":[],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://freidok.uni-freiburg.de/data/1362","outbound_label":"Repository record","outbound_source":"source_url"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Burgard, Wolfram"]},{"key":"dc:creator","label":"Author","values":["Bennewitz, Maren"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:type","label":"Dc Type","values":["DoctoralThesis"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Mehrroboternavigation","Lernen von Bewegungsmustern von Personen","Roboternavigation","Zustandsschaetzung","Vorhersagen von Personenbewegungen","Multi-robot navigation","Learning motion patterns of people","robot navigation","state estimation","predicting the motions of people"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Service robots which act in environments populated by humans have
become very popular in the last few years. A variety of systems exists
which act for example in hospitals, office buildings, department
stores, and museums. Furthermore, several multi-robot systems have
been developed for tasks which can be accomplished more efficiently by
a whole team of robots than just by a single robot. These tasks
include surface cleaning, deliveries, and the exploration of unknown
terrain. Whenever teams of mobile robots are operating in the same
environment their motions have to be coordinated in order to avoid
congestions or collisions. At the same time the robots should perform
their navigation tasks in a minimum amount of time. Thus,
sophisticated path planning techniques are needed that fulfill these
requirements. Since the joint configuration space of the robots is
typically huge and grows exponentially with the number of robots,
existing path planning methods for single robot systems cannot
directly be transferred to multi-robot systems.
Many existing path planning methods for multi-robot systems are
decoupled, which means that they first plan paths for the individual
robots independently. Afterward, they check if the robots would get
too close to each other if the paths were executed. In such a case the
paths are recomputed to avoid these conflicts. Many decoupled methods
assign priorities to the individual robots. These priorities define an
order in which the paths of the robots have to be recomputed. By
computing the path of a robot, the paths of the robots with higher
priority are considered as fixed. This way, the size of the search
space is extremely reduced. Most of the existing prioritized decoupled
methods use a fixed priority scheme (order of the robots). However,
the order in which the paths of the robots are recomputed has a
serious influence on whether a solution can be found at all and on how
efficient the solution is for the overall multi-robot system.
In the first part of this thesis we present an approach which searches
in the space of all priority schemes to find an order of the robots
for which a solution to the path planning problem can be computed.
During the search, we utilize constraints between the priorities of
the robots which are automatically derived from the task
specification. After an appropriate priority scheme has been found,
our technique tries to improve it by using a hill-climbing strategy.
Our search method can be used to find and optimize paths generated by
any prioritized path-planning technique. In several experiments with a
real-robot system as well as in simulation we show that our approach
produces efficient solutions even for difficult path planning
problems.
The second part of this thesis is focused on robots acting in
environments populated by humans. These systems can improve their
behavior if they react appropriately to the activities of the
surrounding people and do not interfere with them. In contrast to a
multi-robot path planning system, the future movements of people are
not known. Therefore, the robots have to be able to detect people
with their sensors, to identify them, and to learn their intentions in
order to be able to make better predictions of their future behavior.
In this thesis we present an approach to learn typical motion patterns
of people from sensor data using the EM algorithm. Furthermore, we
describe how the learned patterns can be used to predict future
movements of the people. Afterward, we explain how this knowledge can
be integrated into the path planning process of a mobile robot.
Finally, we introduce a method which automatically derives Hidden
Markov Models (HMMs) from the learned motion models. These HMMs can be
used by a mobile robot to predict the positions of multiple persons
even when they are outside its field of view. To update the HMMs based
on laser-range data and vision information we apply Joint
Probabilistic Data Association Filters. In practice, the robot becomes
uncertain about the positions of people if it does not observe them
for a long period of time. We therefore propose a decision-theoretic
approach to determine observation actions that are carried out while
the robot is executing its tasks. Practical experiments carried out
with our mobile robot demonstrate that our method is able to learn
typical motion patterns of people, that the navigation behavior of the
robot can be improved by predicting the motions of people based on the
learned motion patterns, that the derived HMMs can be used to reliably
maintain a probabilistic belief about the current positions of
multiple persons even if they are currently not in its field of view,
and that our technique generates effective actions that seriously
reduce the uncertainty in the belief about the positions of people.","Service-Roboter, die in von Menschen bevölkerten Umgebungen
agieren, sind in den letzten Jahren immer populärer geworden. Es
existieren schon eine Reihe von Systemen, die beispielsweise in
Krankenhäusern, Bürogebäuden, Kaufhäusern und Museen
eingesetzt werden. Darüber hinaus sind auch verschiedene
Mehrrobotersysteme entwickelt worden, da einige Aufgaben von einem
Team von Robotern effizienter erledigt werden können
als von einem einzelnen Roboter.
Immer wenn Teams von mobilen Robotern in der selben Umgebung
eingesetzt werden, müssen ihre Bewegungen koordiniert werden, damit
die einzelnen Roboter sich nicht gegenseitig behindern. Außerdem
sollte ihre gemeinsame Aufgabe so schnell wie möglich erledigt werden.
Viele existierende Pfadplanungsmethoden für Mehrrobotersysteme sind
entkoppelt, was bedeutet, dass sie zuerst Pfade für die einzelnen
Roboter unabhängig voneinander planen. Anschließend überprüfen sie, ob
sich die Roboter zu nahe kämen, wenn sie sich entlang dieser Pfade
bewegten. In solchen Fällen werden die Pfade neu berechnet, um diese
Konflikte zu umgehen. Dabei weisen viele entkoppelte Methoden den
einzelnen Robotern Prioritäten zu. Diese geben an, in welcher
Reihenfolge die Pfade neu berechnet werden. Bei der Berechnung des
Pfades für einen Roboter werden dabei die Pfade aller Roboter mit
höherer Priorität als gegeben und unveränderbar angesehen. Auf diese
Weise wird der Suchraum extrem eingeschränkt und die Suche nach einer
Lösung des kombinierten Planungsproblems beschleunigt. Die meisten
existierenden entkoppelten Ansätze benutzen dabei ein festes
Prioritätsschema, d.h. eine feste Ordnung der Roboter. Die
Reihenfolge, in der Pfade berechnet werden, hat jedoch einen großen
Einfluss darauf, ob überhaupt eine Lösung für das kombinierte
Pfadplanungsproblem gefunden werden kann und darauf, wie effizient
diese Lösung für das gesamte Mehrrobotersystem ist.
Im ersten Teil dieser Dissertation stellen wir einen Ansatz vor, der
in dem Raum aller Prioritätsschemata nach einer Ordnung der
Roboter sucht, für die eine Lösung des Pfadplanungsproblems
berechnet werden kann. Dabei nutzt unser Verfahren Constraints zwischen den Prioritäten der Roboter aus, welche
automatisch von der Aufgabenspezifikation abgeleitet werden. Nachdem
ein geeignetes Prioritätsschema gefunden wurde, versucht
unser Verfahren dieses mithilfe einer Hill-Climbing-Strategie zu
verbessern. Unsere Suchmethode kann für beliebige entkoppelte
Planungsysteme eingesetzt werden. In verschiedenen Experimenten
mit einem realen Mehrrobotersystem sowie in Simulationen zeigen
wir, dass unser Verfahren effiziente Lösungen auch für
komplizierte Pfadplanungsprobleme finden kann.
Der zweite Teil dieser Dissertation konzentriert sich auf Roboter, die
in von Menschen genutzten Umgebungen eingesetzt werden. Im Gegensatz
zu einem Mehrrobotersystem sind die zukünftigen Bewegungen von
Menschen aber nicht bekannt. Deswegen müssen die Roboter in der Lage
sein, die Menschen mittels ihrer Sensoren wahrzunehmen, zu
identifizieren und ihre Intentionen zu lernen, damit sie bessere
Vorhersagen über das Verhalten der Menschen machen können. In dieser
Dissertation stellen wir eine Technik vor, die typische
Bewegungsmuster von Personen aus Sensordaten mithilfe des
EM-Algorithmus' lernt. Wir beschreiben außerdem, wie die gelernten
Muster dazu benutzt werden können, um potentielle zukünftige
Bewegungen der Personen vorherzusagen. Anschließend erklären wir, wie
dieses Wissen im Pfadplanungprozess eines mobilen Roboters
berücksichtigt werden kann. Danach führen wir eine Methode ein, die
aus den gelernten Verhaltensmustern automatisch Hidden Markov Modelle
(HMMs) ableitet. Diese HMMs können von einem mobilen Roboter benutzt
werden, um die Positionen von mehreren Personen vorherzusagen, auch
wenn sie außerhalb seines Sichtfelds sind. Um die HMMs mithilfe von
Kamera- und Laserdaten zu aktualisieren, wenden wir Joint
Probabilistic Data Association Filter an. In der Regel wird ein
Roboter unsicher über die Positionen von Personen, wenn er sie längere
Zeit nicht beobachtet. Deswegen untersuchen wir auch, wie
entscheidungstheoretisch geeignete Beobachtungsaktionen bestimmt
werden können, welche ausgeführt werden, während der Roboter seine
sonstigen Aufgaben verrichtet.
Praktische Experimente, die wir mit unserem mobilen Roboter
durchgeführt haben, zeigen, dass unsere Methode typische
Bewegungsmuster von Personen lernen kann, dass das
Navigationsverhalten des Roboters verbessert werden kann, indem er die
gelernten Muster benutzt, um Vorhersagen über die Bewegungen von
Personen zu machen, dass die abgeleiteten HMMs eingesetzt werden
können, um zuverlässig einen probabilistischen Glauben
über die Positionen von mehreren Personen zu behalten, und dass unsere
Technik effektive Aktionen generiert, welche die Unsicherheit des
Roboters über die Positionen von Personen stark reduzieren."]},{"key":"dc:format.medium","label":"Dc Format Medium","values":["application/pdf"]},{"key":"dc:title","label":"Title","values":["Mobile robot navigation in dynamic environments","Navigation mobiler Roboter in dynamischen Umgebungen"]}]}],"canonical_facts":{"dc:contributor":["Burgard, Wolfram"],"dc:creator":["Bennewitz, Maren"],"dc:description.abstract":["Service robots which act in environments populated by humans have
become very popular in the last few years. A variety of systems exists
which act for example in hospitals, office buildings, department
stores, and museums. Furthermore, several multi-robot systems have
been developed for tasks which can be accomplished more efficiently by
a whole team of robots than just by a single robot. These tasks
include surface cleaning, deliveries, and the exploration of unknown
terrain. Whenever teams of mobile robots are operating in the same
environment their motions have to be coordinated in order to avoid
congestions or collisions. At the same time the robots should perform
their navigation tasks in a minimum amount of time. Thus,
sophisticated path planning techniques are needed that fulfill these
requirements. Since the joint configuration space of the robots is
typically huge and grows exponentially with the number of robots,
existing path planning methods for single robot systems cannot
directly be transferred to multi-robot systems.
Many existing path planning methods for multi-robot systems are
decoupled, which means that they first plan paths for the individual
robots independently. Afterward, they check if the robots would get
too close to each other if the paths were executed. In such a case the
paths are recomputed to avoid these conflicts. Many decoupled methods
assign priorities to the individual robots. These priorities define an
order in which the paths of the robots have to be recomputed. By
computing the path of a robot, the paths of the robots with higher
priority are considered as fixed. This way, the size of the search
space is extremely reduced. Most of the existing prioritized decoupled
methods use a fixed priority scheme (order of the robots). However,
the order in which the paths of the robots are recomputed has a
serious influence on whether a solution can be found at all and on how
efficient the solution is for the overall multi-robot system.
In the first part of this thesis we present an approach which searches
in the space of all priority schemes to find an order of the robots
for which a solution to the path planning problem can be computed.
During the search, we utilize constraints between the priorities of
the robots which are automatically derived from the task
specification. After an appropriate priority scheme has been found,
our technique tries to improve it by using a hill-climbing strategy.
Our search method can be used to find and optimize paths generated by
any prioritized path-planning technique. In several experiments with a
real-robot system as well as in simulation we show that our approach
produces efficient solutions even for difficult path planning
problems.
The second part of this thesis is focused on robots acting in
environments populated by humans. These systems can improve their
behavior if they react appropriately to the activities of the
surrounding people and do not interfere with them. In contrast to a
multi-robot path planning system, the future movements of people are
not known. Therefore, the robots have to be able to detect people
with their sensors, to identify them, and to learn their intentions in
order to be able to make better predictions of their future behavior.
In this thesis we present an approach to learn typical motion patterns
of people from sensor data using the EM algorithm. Furthermore, we
describe how the learned patterns can be used to predict future
movements of the people. Afterward, we explain how this knowledge can
be integrated into the path planning process of a mobile robot.
Finally, we introduce a method which automatically derives Hidden
Markov Models (HMMs) from the learned motion models. These HMMs can be
used by a mobile robot to predict the positions of multiple persons
even when they are outside its field of view. To update the HMMs based
on laser-range data and vision information we apply Joint
Probabilistic Data Association Filters. In practice, the robot becomes
uncertain about the positions of people if it does not observe them
for a long period of time. We therefore propose a decision-theoretic
approach to determine observation actions that are carried out while
the robot is executing its tasks. Practical experiments carried out
with our mobile robot demonstrate that our method is able to learn
typical motion patterns of people, that the navigation behavior of the
robot can be improved by predicting the motions of people based on the
learned motion patterns, that the derived HMMs can be used to reliably
maintain a probabilistic belief about the current positions of
multiple persons even if they are currently not in its field of view,
and that our technique generates effective actions that seriously
reduce the uncertainty in the belief about the positions of people.","Service-Roboter, die in von Menschen bevölkerten Umgebungen
agieren, sind in den letzten Jahren immer populärer geworden. Es
existieren schon eine Reihe von Systemen, die beispielsweise in
Krankenhäusern, Bürogebäuden, Kaufhäusern und Museen
eingesetzt werden. Darüber hinaus sind auch verschiedene
Mehrrobotersysteme entwickelt worden, da einige Aufgaben von einem
Team von Robotern effizienter erledigt werden können
als von einem einzelnen Roboter.
Immer wenn Teams von mobilen Robotern in der selben Umgebung
eingesetzt werden, müssen ihre Bewegungen koordiniert werden, damit
die einzelnen Roboter sich nicht gegenseitig behindern. Außerdem
sollte ihre gemeinsame Aufgabe so schnell wie möglich erledigt werden.
Viele existierende Pfadplanungsmethoden für Mehrrobotersysteme sind
entkoppelt, was bedeutet, dass sie zuerst Pfade für die einzelnen
Roboter unabhängig voneinander planen. Anschließend überprüfen sie, ob
sich die Roboter zu nahe kämen, wenn sie sich entlang dieser Pfade
bewegten. In solchen Fällen werden die Pfade neu berechnet, um diese
Konflikte zu umgehen. Dabei weisen viele entkoppelte Methoden den
einzelnen Robotern Prioritäten zu. Diese geben an, in welcher
Reihenfolge die Pfade neu berechnet werden. Bei der Berechnung des
Pfades für einen Roboter werden dabei die Pfade aller Roboter mit
höherer Priorität als gegeben und unveränderbar angesehen. Auf diese
Weise wird der Suchraum extrem eingeschränkt und die Suche nach einer
Lösung des kombinierten Planungsproblems beschleunigt. Die meisten
existierenden entkoppelten Ansätze benutzen dabei ein festes
Prioritätsschema, d.h. eine feste Ordnung der Roboter. Die
Reihenfolge, in der Pfade berechnet werden, hat jedoch einen großen
Einfluss darauf, ob überhaupt eine Lösung für das kombinierte
Pfadplanungsproblem gefunden werden kann und darauf, wie effizient
diese Lösung für das gesamte Mehrrobotersystem ist.
Im ersten Teil dieser Dissertation stellen wir einen Ansatz vor, der
in dem Raum aller Prioritätsschemata nach einer Ordnung der
Roboter sucht, für die eine Lösung des Pfadplanungsproblems
berechnet werden kann. Dabei nutzt unser Verfahren Constraints zwischen den Prioritäten der Roboter aus, welche
automatisch von der Aufgabenspezifikation abgeleitet werden. Nachdem
ein geeignetes Prioritätsschema gefunden wurde, versucht
unser Verfahren dieses mithilfe einer Hill-Climbing-Strategie zu
verbessern. Unsere Suchmethode kann für beliebige entkoppelte
Planungsysteme eingesetzt werden. In verschiedenen Experimenten
mit einem realen Mehrrobotersystem sowie in Simulationen zeigen
wir, dass unser Verfahren effiziente Lösungen auch für
komplizierte Pfadplanungsprobleme finden kann.
Der zweite Teil dieser Dissertation konzentriert sich auf Roboter, die
in von Menschen genutzten Umgebungen eingesetzt werden. Im Gegensatz
zu einem Mehrrobotersystem sind die zukünftigen Bewegungen von
Menschen aber nicht bekannt. Deswegen müssen die Roboter in der Lage
sein, die Menschen mittels ihrer Sensoren wahrzunehmen, zu
identifizieren und ihre Intentionen zu lernen, damit sie bessere
Vorhersagen über das Verhalten der Menschen machen können. In dieser
Dissertation stellen wir eine Technik vor, die typische
Bewegungsmuster von Personen aus Sensordaten mithilfe des
EM-Algorithmus' lernt. Wir beschreiben außerdem, wie die gelernten
Muster dazu benutzt werden können, um potentielle zukünftige
Bewegungen der Personen vorherzusagen. Anschließend erklären wir, wie
dieses Wissen im Pfadplanungprozess eines mobilen Roboters
berücksichtigt werden kann. Danach führen wir eine Methode ein, die
aus den gelernten Verhaltensmustern automatisch Hidden Markov Modelle
(HMMs) ableitet. Diese HMMs können von einem mobilen Roboter benutzt
werden, um die Positionen von mehreren Personen vorherzusagen, auch
wenn sie außerhalb seines Sichtfelds sind. Um die HMMs mithilfe von
Kamera- und Laserdaten zu aktualisieren, wenden wir Joint
Probabilistic Data Association Filter an. In der Regel wird ein
Roboter unsicher über die Positionen von Personen, wenn er sie längere
Zeit nicht beobachtet. Deswegen untersuchen wir auch, wie
entscheidungstheoretisch geeignete Beobachtungsaktionen bestimmt
werden können, welche ausgeführt werden, während der Roboter seine
sonstigen Aufgaben verrichtet.
Praktische Experimente, die wir mit unserem mobilen Roboter
durchgeführt haben, zeigen, dass unsere Methode typische
Bewegungsmuster von Personen lernen kann, dass das
Navigationsverhalten des Roboters verbessert werden kann, indem er die
gelernten Muster benutzt, um Vorhersagen über die Bewegungen von
Personen zu machen, dass die abgeleiteten HMMs eingesetzt werden
können, um zuverlässig einen probabilistischen Glauben
über die Positionen von mehreren Personen zu behalten, und dass unsere
Technik effektive Aktionen generiert, welche die Unsicherheit des
Roboters über die Positionen von Personen stark reduzieren."],"dc:format.medium":["application/pdf"],"dc:subject":["Mehrroboternavigation","Lernen von Bewegungsmustern von Personen","Roboternavigation","Zustandsschaetzung","Vorhersagen von Personenbewegungen","Multi-robot navigation","Learning motion patterns of people","robot navigation","state estimation","predicting the motions of people"],"dc:title":["Mobile robot navigation in dynamic environments","Navigation mobiler Roboter in dynamischen Umgebungen"],"dc:type":["DoctoralThesis"]},"updated_at":"2026-07-24T02:22:10Z"}