hebertmi@EA.USL.EDU (05/27/91)
I'm curious what you guys think of Rodney Brooks. Having no great background in AI (only taken one course on arguments of minds & machines, Hofstadter's Goedel,Escher,Bach) I like his direction.
nagle@well.sf.ca.us (John Nagle) (05/29/91)
What Rod Brooks is really good at is building machinery. His first
six-legged walker was built in three months by him and one graduate student.
It's a complicated and subtle machine, reflecting a detailed understanding
of the sensors and actuators used. For example, he goes into those little
RC servos he uses for leg drive and brings out the error signal, so he
has full force feedback. The little machine has 6 processors (MC68HC11)
linked in what he calls a "token ring" but implemented with a special
serial data path Motorola has on that chip, so no added parts are needed.
This is why he gets results. Contrast this with most of the big robot
projects, where it takes years to build some monster. Robotics is an
experimental branch of engineering, and you have to build machines and
do experimental work.
His architectural ideas are good, and I particularly like his
approch to decomposition. You can go too far in restricting the primtiives
used, and John Connell probably did with the can-picking robot, but
it's useful to have the boundaries explored experimentally. But the
real value of Brooks' work is that he is able to turn his ideas into
working hardware.
John Nagle