Pointers to ftp-able papers (compressed postscript) appear scattered
throughout this page. They and others are available from
ftp.cs.rochester.edu
in directory
pub/papers/systems, .../ai, .../robotics
,
or (if your browser supports it) by following the link.
The students (mostly 2nd year grads but some undergrads and a MS student) in the computer vision class did several projects, usually working in small teams of 2-4. They build and used color segmentation tools in a domain inspired by the Kodak Picture Exchange (KPX). KPX is a way to acquire PhotoCD images over a network via a relational database and user interface, initially for commercial photography. We worked on extracting some of the relevant image properties automatically (such as predominant color, accent color, and empty space where advertising copy could go). The final projects integrated robot control, Jacobian learning (so no hand-eye calibration was needed), and vision. One group used a robot arm, 4-finger manipulator, and two TV cameras to perform pick-and-place tasks on objects that the user clicked in a TV image (this is an example of "deictic control"). A forthcoming Tech. Report will explicate all the various technical details -- watch this spot for an anchor). The other group "herded" a mobile lego tractor along the shortest path through a maze.
Ricardo is actually a systems student, but he did some interesting work
with Brown
that explicated some tradeoffs in communication, parallelism, and performance
in genetic algorithms. The work was implemented on Transputers, as part
of a feasibility test to determine if Transputers were part of the future
of the robot lab. They weren't; a SparcStation
2000 Multiprocessor will be the new main computational and control engine in
the Vision and Robot Lab. Some of the results of
the parallel GA studies are in
systems/93.tr436.parallel_genetic_algorithms.ps.Z.
Jeff should be graduating in August of 1994. His work is in learning robot
control, and he has been building up a theory of motor skill acquisition.
His work can be found in ICRA:
Schneider, J. G. and C. M. Brown. ``Robot Skill Learning, Basis
Functions, and Control Regimes.'' Proc., IEEE Conf. on
Robotics and Automation, May 1993. Also there are some TRs on line
at UR: robotics/92.tr437.robot_skill_learning_and_basis_function_choice.ps.Z
relates early work on learning coefficients of joint velocity basis
functions, and robotics/93.tr458.high_dimension_action_spaces_in_robot_skill_learning.ps.Z describes more
recent work in learning that includes extrapolation from known trajectories.
Bob graduated in 1995, went to work for SGI, and now is at T.J. Watson.
Bob's thesis work is on the interface between large, complex, parallel
systems that embody soft real-time jobs (like reasoning, planning,
world simulation for prediction, truth maintenance...) and hard
real-time constraints (reading sensors, updating controllers...).
As befits a student on the systems and AI interface, he has systems
professors (LeBlanc and Scott) on his committee, and is doing work with
Scott and Kontothanassis on synchronization mechanisms that are affected
by the state of the job scheduler. A not-too-recent description of
Bob's research directions is in
93.PDRTS.Ephor_runtime_environment_for_parallel_intelligent_apps.ps.Z, and a more recent one is
94.RTOSS.An_argument_for_a_runtime_layer_in_sparta_design.ps.Z. A Recent technical report with von Kaenel gives many technical details of a real-world implementation.
Peter finished his 3-2 BS-MS program in 1994, and has for three
years been a pillar of strength for several graduate students to whose
projects he has contributed. His recent work with Ray Rimey was
implementing a dynamic version of Bayes nets to control selective
perception in a world of moving objects, described in
robotics/93.tr466.goal_oriented_dynamic_vision.ps.Z.
Most recently he has been implementing Bob Wisniewski's "shepherding" domain
in the lab; the idea is to track moving lego vehicles visually and to
herd them with a mechanical arm to keep as many of them confined in a
small area as possible.
Jim's thesis was on various aspects of augmented reality, from the use
of affine coordinates for speed (and to avoid calibration) to
implementations of obscuration effect to haptic AR.
See Jim's page for the canonical collection of information
on augmented reality. Jim is a tenured professor at RIT, where he
has been since 1998.
Jim has been working closely with
Kyros Kutulakos.
See Rodrigo's page for a nice MPEG of our automated wheelchair
following him around. He is doing work on pose recovery algorithms as
well.
Amit's thesis was on
is working on Bayesian and probabilistic methods for detection of
interesting properties in comsumer images (such as what the main
subject probably was meant to be). His work has applications in
mobile robot
navigation and multi-sensor fusion.
After a stint out West creatively writing and working for Motorola,
Mike is now in town at the Rochester Institute of Technology.
Mike's thesis was on multi-agent plan creation and organization,
taking place without specific communication or negotiation;
mediated only through
observation, which amounts to plan recognition. Thus his work is
distantly related to the
TRAINS
and TRIPS work of Allen et al.
Chris is working at Equinox corporation; his thesis is on the
properties and applications (tracking, face recognition) of new IR sensors.
He spent the summers of 98 and 99 at SRI
International, working with Bob Bolles and Kurt Konolige, on the
SRI small vision module and associated applications. Summer of 2000
was spent with Equinox Corporation.
Stephen is now at the University of Delaware after receiving his MS
degree here.
In the summer of 2001 Stephen had an internship at IBM Almaden,
and he is interested in teleconference implementations and
related visual algorithms.
A UR Technical Report describing some of his latest results is
available in
Here are the locations of the .gz/.pdf files of the TR
postscript
or
Bo is working on rendering high-quality appearance interactions in
augmented reality (e.g. shadows from virtual onto real objects and
vice-versa, real objects obscuring virtual ones and vice-versa).
He has spent a summer at Siemens Corp. Research doing augmented
reality, and a year putting together an AR system at the Dept. of
Neurosurgery at UR. His page says what we are up to, and
there are two recent TRs:
B. Hu and C. Brown, ``Interactive Indoor Scene
Reconstruction From Image Mosaics Using Cuboid Structure'',
University of Rochester Technical Report 787, August 2002.
B. Hu and C. Brown, ``Acquiring An Environment Map Through Image Mosaicking'',
University of Rochester Technical Report 786, November 2001.
Matt is working on improved techniques for classifying scenes into
semantic categories. He has spent two summers working at Eastman Kodak
Company's Imaging Science and Technology Lab.
As with Singhal, his main
technical supervision comes from Jiebo Luo and Bob Gray of Kodak.
Brown Home
URCS Faculty
URCS Home
Ricardo Bianchini (1995)
Jeff Schneider (1994)
Bob Wisniewski (1995)
Peter von Kaenel (1994)
James Vallino (1997)
Rodrigo Carceroni (1998)
Amit Singhal (2000)
Mike Van Wie (2001)
Chris Eveland (2002)
Stephen Li (2002)
Bo Hu (2003)
Matt Boutell (2003)