ylfink@water.UUCP (10/22/87)
DEPARTMENT OF COMPUTER SCIENCE
UNIVERSITY OF WATERLOO
SEMINAR ACTIVITIES
ARTIFICIAL INTELLIGENCE SEMINAR
- Friday, October 30, 1987
Professor Thea Iberall, visiting the Departments of
Computer Science and Kinesiology, will speak on
``Knowledge-Based Prehension: Capturing Human
Dexterity''.
TIME: 3:30 PM
ROOM: MC 3003
ABSTRACT
A major question facing the development of
sophisticated robotic systems is how to capture the
functionality seen in versatile living systems. One
such versatile function is the dexterity demonstrated
by the human hand under the control of the central
nervous system. As a person reaches to grasp an object,
the hand shapes into a posture suitable for the
interaction. In that the fingers begin forming this
shape as soon as the arm moves, we see this as a
planning problem. In order to develop computational
neural algorithms for grasp planning, we must develop a
likely set of inputs (a task description), outputs (a
hand representation), and a mapping between these.
Since this type of information is currently beyond what
neuroscientists know about the brain, we use a
knowledge-based systems approach to explore potential
representations and mappings. In this talk, I describe
the hierarchical grasp planner that we are developing.
An object database, containing collections of features
about simple shaped objects, is used as input to the
grasp planner. A grasp command points to one of these
objects, and provides the task context for using the
object. Using constraint knowledge bases, the grasp
planner determines reasonable hand postures for the
task.