Page 1 |
Save page Remove page | Previous | 1 of 187 | Next |
|
small (250x250 max)
medium (500x500 max)
large ( > 500x500)
Full Resolution
All (PDF)
|
This page
All
Subset |
PROBABILISTIC MAPS: COMPUTATION AND APPLICATIONS FOR
MULTI-AGENT PROBLEMS
by
Huseyin Hakan Kizilocak
A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Ful¯llment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
ELECTRICAL ENGINEERING
May 2007
Copyright 2007 Huseyin Hakan Kizilocak
Object Description
| Title | Probabilistic maps: computation and applications for multi-agent problems |
| Author | Kizilocak, Huseyin Hakan |
| Author email | caan@caanslaw.com |
| Degree | Doctor of Philosophy |
| Document type | Dissertation |
| Degree program | Electrical Engineering |
| School | Viterbi School of Engineering |
| Date defended/completed | 2002-11-01 |
| Date submitted | 2007 |
| Restricted until | Restricted until 21 May 2009. |
| Date published | 2009-05-21 |
| Advisor (committee chair) | Ioannou, Petros |
| Advisor (committee member) |
Hespanha, Joao P. Sukhatme, Gaurav |
| Abstract | Computation and application of probabilistic maps to multi-agent problems generally consist of generating maps, distributing paths for each agent that allow the group to minimize a cost function subject to constraints, e.g. the length of the path, risk involved with the path, coordinated group behavior, time to arrive at the destination, power consumption, network delay, and bandwidth usage, and deciding to either avoid threats or pursue treasures. In this thesis, we describe a formalism to take care of the general problem of detecting objects with imperfect sensors. This is accomplished by building several probabilistic maps. We develop efficient algorithms to compute probabilistic maps for a large number of static and mobile objects. By a probabilistic map we mean the probability density of the objects' positions, conditioned to noisy sensor measurements.; One of the contributions of this work is that, under suitable assumptions, the joint probability distribution of n objects that lie in a region partitioned into N cells can be approximately determined from an aggregate measurement function that can be represented with memory complexity o(N), regardless of the number of objects n, while the errors in sensors are not Gaussian and sensors are not able to distinguish between individual objects. This is far more compact than an extensive representation of the joint distribution, whose memory complexity is o(N^n). Another contribution here is that algorithms do not have exponential complexity in the number of objects although the positions of multiple objects are not independent of each other. Three main assumptions are: the objects are indistinguishable from the sensor view-point, there is a minimum distance between any two of them, and -- if mobile objects -- their motion is essentially independent from each other.; Innovative applications for this kind of work involve search and rescue operations for disaster sites, e.g. hurricanes, fires, earthquakes, floods, volcanoes, and battlefield strategies, e.g., searching areas for threats or targets, classification of objects detected, pursuit-evasion games, battle damage assessment. One of our applications is to assess the risk of moving from a source to a destination point, exploiting the probabilistic maps, and even to do minimum-risk path planning. Probabilistic maps are constantly computed with the locations of objects by fusing information collected from all the sensors in the region. Additionally, the objects, that we map, may induce danger. Therefore the probabilistic maps may be used for intelligence updates and for minimum risk route planning. They make use of the map-building facility to compute paths for the autonomous mobile robots that minimizes the risk of being in the danger zone. The autonomous mobile robots could be air or ground vehicles depending on the scenario. Another application focuses on the problem where we do not want to avoid the objects that we map, but to pursue. Depending on the particular context, capture may actually mean handling the evader in some particular way, i.e. rescuing it. |
| Keyword | probabilistic maps; multi-agent |
| Language | English |
| Part of collection | University of Southern California dissertations and theses |
| Publisher (of the original version) | University of Southern California |
| Place of publication (of the original version) | Los Angeles, California |
| Publisher (of the digital version) | University of Southern California. Libraries |
| Type | texts |
| Legacy record ID | usctheses-m493 |
| Rights | Kizilocak, Huseyin Hakan |
| Repository name | Libraries, University of Southern California |
| Repository address | Los Angeles, California |
| Repository email | http://www.usc.edu/isd/libraries/services/ask_a_librarian/email/ |
| Filename | etd-kizilocak-20070521 |
| Archival file | uscthesesreloadpub_Volume35/etd-kizilocak-20070521.pdf |
Description
| Title | Page 1 |
| Full text | PROBABILISTIC MAPS: COMPUTATION AND APPLICATIONS FOR MULTI-AGENT PROBLEMS by Huseyin Hakan Kizilocak A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Ful¯llment of the Requirements for the Degree DOCTOR OF PHILOSOPHY ELECTRICAL ENGINEERING May 2007 Copyright 2007 Huseyin Hakan Kizilocak |
Comments
Post a Comment for Page 1
