Futuristic movies portray robots as machines that can react in many of
the same ways as humans.
In reality, current robots rely on a video camera to show them what an
object is.
The future may be more realistic in five years thanks to Yan-Bin Jia,
assistant professor of computer science.
Jia has been awarded a National Science Foundation (NSF) Early CAREER
Award. The five-year, almost $350,000 grant was given to Jia for his proposal,
"Shape Localization, Recognition and Reconstruction through Touch
Sensing."
"Touch sensing has not made any significant contribution to real
applications in factory systems, despite extensive progress on the design
of tactile and force/torque sensors," Jia says. "This may be
attributed to very modest advances in the processing and interpretation
of tactile data.
"My proposal will develop geometric algorithms and software tools
to interpret minimal data necessary and determine the position and orientation
of objects. Why use a camera if you don't have to. The human hand is capable
of exploiting its rich tactile sense and mechanical dexterity to execute
a wide variety of tasks from the brute-force to the delicate types."
While the visual aspects of robotics is well defined, Jia says that touch
sensing is a relatively new in the field. Many non-hardware issues that
surround the topic have not been satisfactorily resolved.
Over the next five years, Jia hopes to gain in-depth knowledge enabling
the robot to infer the geometry and poses of objects from tactile sensing.
He plans to focus on three areas including:
*Object localization from evolution of the contact;
*Tactile recognition of shapes; and,
*Shape reconstruction from tactile data.
"Most of my work is based on geometry and manipulation of robots,"
Jia said. "I would like to enable the robots to have the same dexterity
as a human hand. That way the robot can use other senses besides just
sight.
"Visual sensing is just not enough."
In his research, Jia will be mainly interested in curved 2-D and 3-D shapes.
"This is not only because these shapes often are subjects of maneuver
by the human hand but also because their geometry shares the differential
nature with contact kinematics and dynamics which are responsible for
such maneuvers," he said. "By focusing on curve-shaped geometry,
we can design algorithms to process those curves to special components
or points on the curves, thus helping the robot manipulate the object."
Jia's algorithms will process tactile data, compute geometric invariants
and register them onto the object's surface.
"My approach to model-based tactile recognition will build on localization
and invariant-based 'geometric hashing,'" he said.
Jia plans on combining techniques from curvature estimation, kinematics-based
recovery of local patches, and global fitting. His research will encompass
issues across several of the most active areas of robotics including touch
sensing, parts orienting, shape recognition and reconstruction, grasping,
dexterous manipulation, and haptics. The work will build on his background
in differential and computational geometry, algorithms, optimization,
numerical methods, nonlinear control, mechanics, sensor technology, artificial
intelligence, and robot programming.
Jia predicts that the project will have industrial applications.
"Factory assembly has to design all-purpose robots that rely on visual
sensing," he said. "If you can combine visual with touch sensing,
that will increase the type of applications the robot can perform."
Each NSF CAREER Award must also have an educational component to it. Jia
plans to train interdisciplinary engineers and researchers with a focus
on robotics and geometric computing, A new graduate course would introduce
a collection of mathematical and computational tools that have wide applications
in computer science and engineering. A new undergraduate robot laboratory
course will foster students' hands-on skills with robot sensing and manipulation
as well as demonstrate results from Jia's research project.
Around LAS
March 11-31, 2002
|
