Re: DARPA Grand Challenge

On Wed, 21 Sep 2005 13:33:12 -0400, HMS Beagle <bgates@xxxxxxxxxxxxx>
wrote:

>The lesson here is that there is MASSIVE PROCESSING going on in the
>retina alone. People who work in vision have actually created robots
>that use a giant grid of cells to perform this edge-detection
>mechanism. The cells are very very simple, but there are thousands of
>them. Google "neuromorphic engineering".

Of course, if you use as many light detectors as the human eye, you're
going to have massive processing. But I don't think a robot needs to
have either the resolution, the contrast or the color discrimination
capability of the human eye to be very useful. After all, there are
insects and other animals that can navigate perfectly well with visual
systems that are orders of magnitude less sophisticated.

IMO, the visual resolution needed to navigate a course such as DARPA's
Grand Challenge course, need not be greater than 10,000 receptors. And
a monochrome 16 to 1 contrast ratio would be more than adequate.
What's much more important is the subsequent processing. I'm thinking
of such things as prediction, memory sequencing, concept formation,
attention, adaptation, etc...

On a related tangent, even though there are a lot of cells in the
retina, the processing per cell, as you pointed out, is very simple. A
fast-response retina can be implemented entirely in hardware obviating
the need to do it in software.

>So yet again, we have this issue of linear processing vs. massive
>parallelism of simple units. It turns out, Rodney Brooks'
>subsumption architecture was an attempt to add more parallelism into
>our systems, not so much to "remove heavy processing". It seems that
>heavy processing is something we are going to have to live with one
>way or another. Even the human body is seen doing massive
>"calculations".. such as in the eye as above.

Yes but doesn't have to be nearly as massive as the pundits have
claimed. We can have extremely useful artificial brains/robots with
less than one hundredth of the capactiy of the human brain. Take a
good look at what lower animals can do with what they have. A
honeybee, for example, has an amazingly sophisticated and complex
behavioral repertoire and yet it has only about 1 million neurons!

Louis Savain

Why Software Is Bad and What We Can Do to Fix It:
http://www.rebelscience.org/Cosas/Reliability.htm
.

Relevant Pages

  • Re: Reply to Wolf
    ... probably thinking of the so called "simple cells" of the primary visual ... You might enjoy Helga Kolb's survey article: "How The Retina Works" ... Here is the kicker - individual neurons in V1 can be made to respond to ... It amounts to responding to a single stimulus dimension. ...
    (comp.ai.philosophy)
  • Re: Ben G on reinforcement-learning and the wirehead problem
    ... ``In the retinal ganglion cells there are two types of response, ... centre of the receptive field causes the firing rate to increase. ... That's not edge detection Tim. ... Uh - most edge-detection happens in the retina. ...
    (comp.ai.philosophy)
  • Re: Mission Impossible
    ... But also we do not know yet how eye- brain system works, ... activity that regulate eye growth, and direct response of these cells ... "Axial elongation is triggered by"hyperopic blur on peripheral retina" ... Same work suggests which cells initiate invoke signals for eye grows: ...
    (sci.med.vision)
  • Re: So, Im pretty sure I saw my own blood cells.
    ... >>>seeing an image of the blood vessels in our retina, ... >>>seeing individual blood cells, though. ... >> cells flowing through your retinal circulation. ... If so, look up "floaters". ...
    (talk.origins)
  • Re: Comparitive visibility of red and green lights
    ... It has been suggested that the human eye uses the difference in image ... intensity found at the retina between red and blue light to focus ... This would mean that the dark focus adapted eye would tend to see red ...
    (sci.optics)