On Research:
Teaching Robots to Be Smarter


Ken Keatley
---------------------------------
Homewood News and Information

     Robots that swim along the ocean floor in pursuit of elusive
plankton, and those that spot-weld cars on a high-speed assembly
line, are plagued by similar problems: how to get the job done
faster and more accurately. For researchers, the fundamental
question is, How do we make robots smarter?

     Louis Whitcomb, associate professor of mechanical
engineering in the School of Engineering, is working to find
solutions to the control and dynamics issues that hold back new
or advanced uses of robots.

     "The goal is to develop robots that are intelligent in
learning about themselves and their environment, and can use that
information to improve their performance," Whitcomb says.

     While there is much more to be learned, Whitcomb and the
robots he's helped develop are literally making great strides. He
was the architect of a new control system for JASON and ARGO II,
the Woods Hole Oceanographic Institution's remotely operated
underwater vehicles. Last summer, ARGO II was successfully
deployed 3,500 meters beneath the surface of the Atlantic Ocean
to investigate hydrothermal vents.

     The state-of-the-art robot, linked by fiber-optic cable to a
control room in a ship on the surface, permitted precise data
collection in a complex environment. Such instruments as an
electronic still camera (which took 30,362 photos for an
oceanographic survey) and sonars were operated via joystick from
the mother ship floating over two miles above the ocean floor.

     "We had to hover ARGO II just off the bottom, and position
it back and forth over a vent," Whitcomb explains. "We were able
to make the vehicle go where it was supposed to go, enabling the
robot to bring back data that will help geologists unravel the
mystery of hydrothermal vent formation."

     This summer, Whitcomb will again collaborate with Woods
Hole scientists in developing a prototype control system
that will enable an underwater robot equipped with two cameras to
track the movements of plankton for an extended period of time.

     On land, Whitcomb has been equally successful. He is a
consultant to a number of the world's leading manufacturers of
industrial robots, and recently spent a year in Japan
collaborating with engineers at the University of Tokyo and at
the Toshiba Corp. Research and Development Laboratory. He
formerly worked as an R&D engineer at GMFanuc Robotics Corp., the
largest robotics vendor in North America.

     "For a lot of operations in the body shop of an automobile
factory--like arc and spot welding, or painting--robots are the
right way to do the job. They are especially useful in replacing
human workers in jobs that are dirty, dull or dangerous,"
Whitcomb explains.

     Nevertheless, earlier generations of robots have been less
successful when the application--like deburring of rough edges,
or cutting patterns in metal surfaces--requires precision in an
ever-changing environment. Whitcomb is developing and applying
model-based adaptive control techniques that compensate for such
forces as gravity and inertia, and result in high-performance
systems.

     "You can write a mathematically precise control algorithm
that gives the machine a chance to adapt, to interact dynamically
with its environment," he continues.

     While Whitcomb has seen many of his control systems put into
place on the assembly line and beneath the ocean, he is well
aware of the challenges ahead.

     "It will be a long time before robots are as smart as a
6-month-old child," he says.

     Whitcomb, who joined the Hopkins faculty last fall, is part
of a burgeoning group of robotics engineering researchers. Others
are fellow mechanical engineer Gregory Chirikjian and biomedical
engineer Reza Shadmehr.