Biologically Inspired Control of Intelligent Machines
George Bekey,
USC
12:00 Wed. Oct. 11 in 159 Mechanical Engineering
Living organisms are highly adaptive to changing environments.
By contrast, industrial robots and other machines are best
suited for highly structured and predictable environments, where
control can be based on accurate mathematical models of both
machine and environment. In this talk we present recent and
current work at USC on the control of autonomous systems capable
of operating in unstructured environments independent of human
control, using biological inspiration rather than classical
control theory. We have built wheeled, legged and flying robots
based on these ideas.
The first quadruped walking machine entirely under computer
control was built at USC in the early 1960's, and called the
``Phony Pony.'' Recently we have built both quadruped and hexapod
walking machines. A unique feature of the hexapod was that its
gait patterns were ``evolved,'' using a genetic algorithm to set
weights in a neural network, rather than being programmed. One
of our wheeled robots uses reinforcement learning to avoid
obstacles. We have also constructed autonomous helicopters,
which are controlled using a behavioral paradigm. Unlike
terrestrial robots, the helicopter must constantly make correct
decisions to prevent it from crashing. The architecture of the
controllers for these machines will be described, with an
emphasis on the advantages and difficulties of actually building
robots, rather than only simulating them. Videotapes will be
used to illustrate these weird and wonderful creatures.