DPRG
DPRG List  



[DPRG] Is subsumption old hat? .... and who cares what the academics and doing?

Subject: [DPRG] Is subsumption old hat? .... and who cares what the academics and doing?
From: Huff, Brian L bhuff at uta.edu
Date: Fri Sep 7 21:50:57 CDT 2007

Dear All, and you too Randy...

First I side with the "Hobbyists" not because Subsumption is the end
all, be all, but because they take what is available to them, figure out
what works and does not work for their environment/problem and they
implement it.  The simpler and cheaper, the better.  

I have been in academia for about 20 years and have been passionate
about industrial robotics and for the last couple of years, autonomous
systems.  I am not sure that I know what "the next big thing" is.  I
also want to state that the systems I put together in my spare time can
not hold a candle to those on display at the typical DPRG event.  

Maybe the Hobbyist Community is not aware of the next great thing
because there has not been another Paradigm shift like Subsumption back
in the mid-80's that has proven helpful to the robot developer, working
on small computing platforms, who is trying to get something to work in
the real world.  Or maybe they are not willing to wade through the
typical BS (not Bachelors of Science) put out by academics on robotics
to try and identify the pearls of knowledge that sometime can be found. 

The current wisdom in Robotics seems to be uncommonly practical, "Use a
Deliberative Control Strategy When it makes sense and revert to a
Behavior-Based approach when your well laid plans blow up in your face."
This hybrid architecture, in my humble opinion, is a great approach.
The only problem is, this practical "use the best of both worlds"
approach is not very sexy for those trying to make a name for themselves
in the world of robotics (common sense does not seem to help with
obtaining tenure).

This hybrid approach has also not helped the typical hobbyist because
our little robots typically do not have the computing horsepower or the
software infrastructure to support the historic AI Deliberative Planning
/ World Modeling approach to robotics.  This is changing however with
smaller, better, cheaper controller boards and sensors hitting the
market all the time.  The problem is, the problem space covered by
academics with large computing platforms and substantial research
budgets has become scorched earth.  The existence proofs now exist.  We
have proven that a humanoid robot can walk up stairs, why should we
spend millions of dollars more to get it to do something useful now that
it is there?  Being the second one up the stairs (even at 1/10 of the
price) really doesn't matter much either, unless of course it allows you
to do something cheaper or better than you could do before.  Then it
stops being a robot and becomes a hammer.

Hobbyists will pay attention to academic research when academic research
produces something of value to the Hobbyist community.  Subsumption
provided a model for small computing platforms to deal with events in
unstructured environments in a manner that was never possible using
deterministic procedural programming techniques.  That was a huge tool
for the robot builder.  Combining this with the historic
Sense-Model/Represent-Reason-Act paradigm found in traditional AI
robotics is simply not that helpful to someone trying to get his robot
to scoot across his living room without getting stuck under the coffee
table.

You know... I do know what the next big thing is in robotics.  It is
that we are no longer just playing with these little toys in research
labs at MIT and Stanford but we are starting to field these things in
the hundreds in places like Iraq and Afghanistan.  The next big thing is
that robotics is a technology who's time has come.  The basic building
blocks are becoming cheap enough for the average school kid to own and
play with.  These systems are now saving lives (and assisting to take
lives) on the battlefield today.  But if you take a good look at the
robotic tools in the hands of today's war fighter they more resemble R/C
toys on steroids than the high-powered "thinking machines" promoted in
academic research.   So just maybe it should be the academics who should
be pay attention to the hobbyist and not the other way around.

Brian Huff
    
          

-----Original Message-----
From: dprglist-bounces at dprg.org [mailto:dprglist-bounces at dprg.org] On
Behalf Of Randy M. Dumse
Sent: Friday, September 07, 2007 7:44 PM
To: dprglist at dprg.org
Subject: RE: [DPRG] Is subsumption old hat?

Jeff Koenig stepped up and said: Monday, August 27, 2007 12:54
PM
> I'm betting few of us have the means to produce a complex
board, 
> burn FPGA's or CPLD's, and tweak a compiler to use the board's
> memory map.

But some of us do.

And generally, if there's demand for something, someone steps up
and offers a product. I was one of those who did just that some
~25+ years ago. I would do it again. But it really seems to me
at the moment, offering a 40-MIPs DSP was something to fresh for
the market.

> I really like subsumption. I don't know what's presently in 
> vogue, but for small microcontrollers and real-life robots, 
> subsumption has worked well for me.

I really think you're making my point here Jeff. If you don't
know what is in vogue, how do you know if it is useful on small
microcontrollers or not?

This is not really meant as criticism, and please don't take it
personally. I would be in the same boat, except after finishing
Brooks Flesh and Machine, I read Arkin's Behavior-Based
Robotics. That gave me a feel for advances. Then I started
reading Murphy's Introduction to AI Robotics I got the
impression lots had happened since 1992. Having not yet finished
the book, I am still wondering how big these advances really
are. 

But what I find most interesting is that it seems the hobby
community is largely unaware of any advances. So I wonder
advances, there or not? And if there, how big? Or don't they
much matter. Murphy thinks so. Or she wouldn't have made the
comment about 1992. So there's some kind of interesting
disconnect from the research community and the hobby community.

Randy
www.newmicros.com


_______________________________________________
DPRGlist mailing list
DPRGlist at dprg.org
http://list.dprg.org/mailman/listinfo/dprglist

More information about the DPRG mailing list