|
|
 |
Introduction
The Robotics Seminars are intended to appeal to a wide technical audience.
A variety of talks are presented that are of interest to computer scientists,
electrical engineers, mechanical engineers and others in the fields of mobile robotics,
manipulation, and human/robot interaction. The seminars are free and open to the public.
|
| |
|
Fall 2008
Wednesday Sep 3, 1:00 pm - 2:00 pm in
Walter Library (DTC) 402
Speaker: Dr. Gian Luca Mariottini
Title: Image-based Robot Control: the Multiple View Geometry Approach
Abstract: In the last few years there has been an increasing interest for robotic
applications in the fields of medicine, industry or space. Vision could be of great interest in all
of these fields: cameras are less expensive than lasers and sonars and provide non-contact measurements
of the surrounding environment. Their versatility continues to stimulate many robotics research fields
(localization and mapping, multirobot, bio-inspired navigation, etc.) also motivated, e.g. by the
achievements in the computer vision research community. However, vision is still not used as the main
(or alone) on-board sensor, but jointly with others (e.g. inertial measurement units, lasers, etc.).
In this seminar, I will show how the computer vision paradigms can be used for the real-time navigation
of robots towards a goal configuration, specified exclusively by its reference view. In particular,
I will address the problem of using the algebraic theory of multiple-view geometry for vision-based robot
navigation (or visual servoing) using only visual information provided by an on-board pinhole or omnidirectional cameras.
Using this approach, typical problems such as local minima or Jacobian singularities, typically encountered
in image-based visual servoing schemes, are here eliminated. Also, no metrical knowledge of the 3-D scene is
needed. Moreover, and differently from the well-known 2-1/2-D techniques, our algorithms do not estimate the
relative robot/camera displacement (e.g. via homography matrix decomposition). This process, in fact, would
require manual intervention and could also increase the sensitivity to image noise. Asymptotic convergence to
the desired robot configuration has been proved, also in the case of unknown focal length. Extensive
simulative and experimental results will be presented.
About the speaker: Gian Luca Mariottini (1977) received his Master Degree cum laude in
Computer Science in 2002, and the Ph.D. Degree in 2006 from the University of Siena. His Ph.D. thesis is
on vision-based robot control, with particular emphasis on the control and sensing aspects (pinhole and
omnidirectional cameras). He collaborated with Prof. Giuseppe Oriolo of the University of Rome "La Sapienza"
and has been collaborating with Prof. Domenico Prattichizzo (University of Siena, ITALY). He taught Robotics
and Computer Vision courses for undergraduate and master students and tutored more than 30 of them in their
final theses. During his PhD, he has been a visiting scientist at the University of Pennsylvania, Philadelphia PA,
and he has been collaborating with K. Daniilidis and G.J. Pappas on nonlinear localizability problems for
multi-robot formations with on-board omnidirectional cameras. He has ongoing collaborations also with
the physicians of the Psychological Dep. of the "Policlinico Le Scotte", University of Siena, for studies related
to eye-tracking and transcranial magnetic stimulation effects on human emotive status. In 2007-2008 he moved to
the Georgia Institute of Technology as a Post-Doctoral fellow, to work on the use
of visual sensors for visually impaired people navigation and localization in city-like environment (SWAN Project).
|
| |
|
