Re: forests on orbit

It just ain't that easy, Ian.

If it is for you and you can make all this stuff work with 100%
reliability in the real world, I can probably get you a job where I
work.

Ian Parker <ianparker2@xxxxxxxxx> wrote:

:On 13 Feb, 21:04, Willie.Moo...@xxxxxxxxx wrote:
:> You are avoiding the central issue I raised.  Your jig that you put on
:> the loo seat has to be fitted by a human.  HOW DID HE OR SHE DO THAT?
:> We do not know!  That's the simple answer.  Until we do know - even on
:> a theoretical level - its premature to design whole systems around
:> solving a problem that we don't even know how to state let alone
:> solve.
:>
:> You said a robot is a general purpose jig.  Okay.  That's a theory.
:> You talked about finding where something was.  How do you do that?
:> Recognizing what something was?  Didn't even address that.  Figuring
:> out what to do about it.  Translating things?  We know how to do
:> coordinate transformations - people have tried that -it doesn't work
:> the way you think.  BUILD  IT AND SEE!  Marvin Minsky thought we'd
:> have human level robots by the 1990s.  Didn't happen.  We learned that
:> we don't know enough about the problem to even THINK about it.
:>
:> Look at y challenges.  Get a GP robot to pick parts out of a bucket of
:> parts and put them together in some simple fashion - DO IT!  When you
:> try to do it - you find that you don't even understand the problem -
:> though you think you do.  THAT'S THE PROBLEM!   We EMBODY a thing that
:> SEEMS so simple to us, but BECAUSE we embody that thing, we CANNOT
:> understand it - and we need to understand it in order to build a
:> machine to do it.
:>
:> I'm not an unreasonable person.  SHOW ME - a machine that embodies the
:> same mechanical and spatial skills we embody, even on a primitive
:> scale - show me a THEORY of how to build such a machine that addressed
:> the unresolved issues developed by roboticists over the past 50 years
:> - THEN I will say, sure, I'll plan for this sort of thing.
:
:I have tried not to beg the question. In this posting I will be
:discussing pattern recognition and optical flow only. I am assuming
:that once an object is found it will be possible to take its
:coordinates and do something with it.
:http://en.wikipedia.org/wiki/Pattern_recognition
:
:This is Wikipaedia giving an overview of pattern recognition.
:
:http://www.stats.ox.ac.uk/~ripley/PRbook/ A book on Pattern
:recognition.
:
:I have I think mentioned competitions before. The DARPA Urban
:Challenge is perhaps descriptive of the state of the art in pattern
:recognition. If I was given a job by Marriott Hotels to look at room
:service DARPA UC is among the first places I would look. A car CAN
:navigate in a urban environment. DARPA's car (in a suitable form)
:could go into a hotel room and into the loo. This as I think I have
:implied earlier is 90% of the battle. OK the DARPA car is quite heavy
:and has a lot of computational equipment. However ASICs were not used
:and they could reduce the size considerably. Also mullticored
:processors are coming out and the box in the DARPA car could be made
:to fir on a board.
:
:http://www.google.co.uk/search?hl=en&q=DARPA+urban+challenge&meta=
:DARPA Urban challenge. Driving in a urban environment to me represents
:the current state of art.
:
:The first thing you need to do to find where an object is is to
:perform edge detection. This will reduce the number of image points to
:a convenient size.
:
:http://en.wikipedia.org/wiki/Edge_detection Methods of edge detection
:
:http://www.pages.drexel.edu/~weg22/edge.html Edge detection tutorial
:
:If you have a robot that moves, suppose you have a chambermaid's
:trolley which we make automotive, you will need to supplement edge
:detection with optical flow measurement.
:
:http://www.google.co.uk/search?hl=en&q=optical+flow+detection&meta=
:Optical flow detection.
:Optical flow will tell us how far away a line on the edge detection
:algorithm is. If you have a drawing of your loo seat in AutoCad or
:ProEngineer you are now in a position to fit your shape. Now a loo
:seat is a coplanar object. All edges therefore will be coplanar and
:also smooth. We have to find closed paths whose curvature is
:continuous. These closed paths are then matched to our object. As we
:know how far away the object is we therefore know the absolute size
:and orientation - bingo!
:
:If you want to make beds you again have to look for the basic shape.
:If we are in a hotel it should be remembered that the architecture
:will be coded into a 3D drawing passage, so we can navigate using dead
:reckoning with updates.
:
:One of the bees I have in my bonnet is that there is, as of now, no
:universal robot operating system. The software tends to be written
:from scratch in every case and this leads to inefficiency. What is in
:fact needed in the pattern recognition part of robotics is a package
:of subroutines which will allow recognition of ProEngineer drawings.
:
:
: - Ian Parker
.

Quantcast