Robotics
-
Urban Search and Rescue [USAR]
Each year, the American Association for Artificial
Intelligence holds a robot competition. For the past 3
years the competition has included an Urban Search and
Rescue Task. NIST [Nat'l Inst. of Standards and
Technology] has developed a "standard" course for this
competition that is extremely challenging. This course
is currently setup at NIST in Maryland and is
available for research groups to use at any time.
The challenges in USAR are map building, detection of
important locations, navigation in a cluttered,
unknown environment using multiple sensors, and path
planning to an exit. In addition, I would like to
explore semi-autonomous solutions where one person is
able to control multiple roots through a sophisticated
graphical interface.
-
Navigation and Interaction in a Human Environment
That past few years we have developed successful
platforms for serving in a crowded
environment. Aspects of this problem include visual
analysis, speech recognition, facial animation,
navigation, and localization. I would like to build
upon our knowledge and create a robot that could
generally navigate and interact appropriately in an
environment such as Hicks, the Kohlberg lounge area,
and Sharples.
-
Navigating Hicks with one or more robots
Having robots navigate and map buildings is necessary
for them to start to do useful things like collecting
and distributing mail, looking for and finding people,
and other useful tasks (be creative). Navigating Hicks
offers many challenges, including forcing the robot to
be able to find and use the elevator.
in 1999, Carl Wellington and Roger Bock designed and
built an adjustable arm and finger for pushing
elevator buttons. In 2000, Brandon Rutter completed
the work and got the robot to push an elevator button
using visual servoing. The tasks ahead include getting
the robot to find the elevator, navigate in and out of
it, and improved visual servoing to find the elevator
button. All of our robots now have wireless
communication capability, which will make development
and testing easier and faster.
It is likely we could have two or more robots doing
this task, so cooperation is possible. An interesting
variation of this project involves having the robots
try to play hide and go seek, which would be fun to
program and to watch.
-
Mapping and localization
Mapping and localization is a difficult problem in
robotics. Using multiple sensors and building maps we
may be able to do a better job than using a single
sensor or open-loop control. We can also try to
engineer the environment by dropping our own markers
(RF markers, or visual markers) and letting the robot
use them to localize itself.
-
Any number of students could work on various aspects of these
projects. They involve some hardware work (maintaining the
robots and components) but will be mostly programming and
testing.
|
Mechanical Design and Embedded Control Systems
-
Development of a Calibrated Imaging & Simulation Lab
In 1999 Ani Hsieh began the development of a
calibrated imaging and simulation lab by designing and
constructing a 6 DOF positioning device capable of
positioning a 2kg load within a 1m workspace. RJ
Berger continued this project in 2002 and wrote a
graphical user interface for controlling the
positioner. The next step in this project is to
calibrate and test the device for accuracy, and to
collect some calibrated image data sets. These data
sets are useful for research into stereo vision and
motion algorithms. In addition, we can work on
improving the control algorithms for the positioner,
further develop the GUI interface, and use it in
simulations and tracking tasks such as locating and
moving towards people in the lab.
One or two people could work on this task, which could
involve GUI development, control theory, mechanical
design, computer vision, or any combination of the
these areas.
|
Computer Vision
-
Developing an advanced segmentation system for content-based
image retrieval
One of my major interests is intelligent segmentation
of images: dividing an image up into regions that
correspond closely to the objects in the image. This
is an extremely challenging problem (try defining
"object"), but we have made some good progress and I
want to continue this work.
At its core, this project will involve using your
knowledge of physics, engineering, and mathematics to
devise ways to define and test whether a given
boundary is between two objects of just divides a
single object. This is fundamental computer vision at
its best, so come talk with me if you might be
interested. There are lots of things to explore and
you should be able to build upon my work to develop a
fairly sophisticated approach to the project during
the course of the semester.
One or two students would be appropriate for this
project, which will involve most programming and
probably some numerical receipes-type algorithms.
Developing a Virtual Chalkboard
The goal of this project is to develop a virtual
chalkboard system. It would use visual and auditory
input to allow a professor to generate and manipulate
3-D models in real time using gestures and spoken
language. The project will involve tracking an object,
interfacing with a 3D virtual environment, gesture
recognition, and speech recognition.
Robot vision: mapping what the robot sees and identifying
landmarks
This is a standard problem in robotics/computer
vision. Typically, roboticists use easy-to-spot
landmarks like bar codes or brightly colored
patterns. For a robot to navigate successfully in a
non-specialized environment, it must be able to find
and remember natural landmarks.
One or two students could work on this project, which
will involve mostly programming and testing out
different object recognition/detection
techniques.
|
Computer graphics
-
Stained glass
I'm interested in developing realistic looking stained
glass images, both from modeled scenes and from
existing images. A further step will involve animating
the scenes in a realistic manner.
|
Pattern Recognition and Data Mining
-
Pattern recognition & segmentation for automatic typology
generation with hydrogeological data
This is a project that Dr. Maxwell is continuing to
work on with an international group. The LOICZ data
set is a hydrogeologic data set that contains data
from coastlines around the world. Hydrogeologists are
currently using this data to develop a typology of
coastlines (is Antarctica like New England?).
We are using computer vision/pattern recognition
segmentation and clustering techniques to do a
data-driven classification of the coastlines. There
are multiple computer vision techniques which are
appropriate to this task, and an E90 project could
focus on applying 2-3 of them and comparing the
results. This is a "hot" problem in hydrogeology and
there is a high probability of getting a paper
published from this project.
In 1999 Jennifer Vander Veer developed a GUI for the
LOICZ clustering tools. The hydrogeologists loved it,
Casey Smith implemented version II in the summer of
2000 and has continued that work in 2001. More
development is needed in the science, and you should
see results fairly quickly. This project will also
involve developing new visualization techniques in
order to more easily assess the data and the resulting
clusters/typology.
One or two students would be appropriate for this
task, which will involve C programming, web
programming and visualization of large scientific data
sets.
|
|
|