Curious?????

From: Ed Minchau
To:
Sent: Monday, April 16, 2001 2:03 PM
Subject: [spacesettlers] Re: Curious?????

[snikt]
> Robots controlled by expert systems do not work in unstructured
> environments, but a robot with a control system based on fuzzy logic,
> neural nets, and genetic algorithms can be _taught_ to work in poorly-
> defined environments. I have been working on an autonomous robotics
> project for ten years, and I should have a proof-of-concept model
> ready later this year (2 fully autonomous serpentine robots, also
> capable of teleoperated semi-autonomous mode).
>

I don't now your research and I therefore I can emit no opinion about it.
However, since you mentioned connectionist/evolutionary approaches... This
fuss about neural nets, fuzzy logic and genetic algorithms has lasted for
the past 20 years and so far I've seen those techniques applied just to
toyish academic problems, when it comes to Robotics: robots that run thourgh
rooms avoiding boxes, robots that follow tape patterns glued in the walls,
and so far. I have no knowledge of robots in real industrial applications
(such as mining) using those connectionist/evolutionary approaches to
actively manipulate unstructured environents without human supervision. (If
you do know something like that, I would greatly appreciate any references
that you ca present.) And although I agree that in *theory* that can be
accomplished, the near stagnation in that field makes me doubt that this
accomplishment will occur in a foreseeable future.

Therefore, I would invest in sending human miners to asteroids. They could
operate telepresence rigs to do the excavations (which would improve their
safety), but use of robotics would be limited to that. Conversely, you can
drag an asteoroid to HEEO and tele-mine it from Earth itself.

> :) ed
>

;) lcio

P.S.: I hope that you are not thinking that I'm one of those hard AI types
who hate neural nets and evolutionary computing... In fact, I think that
those techniques are great and have many successful domains of real-world
application; I've been working with them since 1993. I just think that
Robotics, from the commercial/industrial standpoint, is not one of those
successful domains...