Subject:Re: Mark Tilden and BEAM -- just a bunch of hype?
Date:Fri, 12 Mar 1999 21:54:10 GMT
From:Don Roy 
Organization:Personal Robotics
Newsgroups:comp.robotics.misc
References:1 , 2

Adam wrote:
> 
> I've had a bit of a look at some BEAM circuit diagrams and there
> doesn't really seem to be anything remarkable about them. As far as I
> can see, the entire circuit is really just a serious of pulses (albeit
> analog) driving some transistors and integrators.
> 
> I'm confused as to how these things can "learn" or "adapt". It would
> seem to me that it is more a case of if a motor stalls or is blocked
> by an obstacle then more/less current flows through that point in the
> circuit. The change in current causes the other capacitors etc to
> charge differently, which means that the next series of movements will
> be different. Eventually the change is enough to circumvent whatever
> problem is causing the motor to stop in the first place.
> 
> This are really just my thoughts - I am by no means an electronic nor
> robotic expert. But IMHO there is no intelligence as such to be
> perceived in these robots. Please don't think I'm knocking Mark or his
> bots in any way - I think they're good for what they are - but I think
> I'll be sticking to my uP to do my work.
> 
> Comments?
> Adam

As it happens, I don't find the _building_ of BEAM robots to be
particularly intriguing either. However, I do appreciate what the BEAM
folks can accomplish without the use of a uP. In my humble view, they
are exploring the area of reflexive behavioral response to environmental
stimulus, which is a valid and useful direction.

In fact, properly integrated in support of uP operations, there is a lot
to be gained in taking the best of both approaches. A good example of
this is Rod Brooks' description of how a hexapod robot walks when using
a subsumption architecture (MIT AI Memo 1091). See:

http://www.ai.mit.edu/people/brooks/papers/AIM-1091.pdf

Following a signal from a central source, the remainder of an individual
leg's actions are entirely reflexive... i.e.: the bot has no choice but
to walk. Point is that the central controller is not tied up in handling
the minutae. Whether the reflexive actions are handled by subprocessors,
analog circuits or gremlins is relatively unimportant.

Some further ideas along this line, including the potential emergence of
"complex" behavior from simple circuits, are explored in Braitenberg's
"Vehicles" book... recommended reading.

While there might appear to be some "us versus them" division, I think
it would be a dreadful mistake for anyone to dismiss the BEAM approach
out of hand.
Just my $0.02 ...Don
--
________________________________________________
FirstBOT Mobile Robot - Free plans and info at:
       http://www.personalrobotics.net
Personal Robotics: The Second Cambrian Explosion