[Prev][Next][Index][Thread]

Re: Steam

We've got some code that can see about 60% of the walls
from a home grown BW CMOS camera sensor at 150 mm
above the maze surface. Once the camera is calibrated we
can find the walls in under a second using a Hitachi SH-3
(80 MIPS). The camera calibration procedure takes about
10 minutes. This can be improved with an SH-4 (200 MIPS)
but is still not practical. The trick as I see it is to pre-calibrate
the more complex camera parameters before the run and then
do a quick re-calibrate when placed in the maze for the final
run. You then need to drop the sensor down low and use it as
a more conventional range-to-wall sensor.

We haven't got this final stage together yet - all my students
prefer robot soccer :-). I don't believe it will be as fast as
CUQEE III, but it will be extremely cool anyway.

BTW, there used to be a prize for Smartest mouse which
was the mouse that had the best first run to the centre, i.e.
fastest initial explore.

Gordon Wyeth
Department of Computer Science and Electrical Engineering
University of Queensland     4072 AUSTRALIA
Phone: +61 7 3365 3770         Fax: +61 7 3365 4999
wyeth@csee.uq.edu.au         http://www.elec.uq.edu.au/~wyeth
----- Original Message -----
From: John Wright <j.wright@ieee.org>
To: A Johnstone <adrian@dcs.rhbnc.ac.uk>; John Billingsley
<johnbill@usq.edu.au>
Cc: <micromouse@dcs.rhbnc.ac.uk>; <micromousers@ucdavis.edu>
Sent: Saturday, 1 July 2000 0:19
Subject: RE: Steam


> Adrian,
>
> Again, not that new of an idea, but technology has advanced.  A mouse was
> run in Southern Region 6 with a CCD on a fishing pole.  It took a snapshot
> of the maze and then pulled the pole in and just sat in the start cell
> processing the image.  After 5 minutes of nothing the judges pulled it off
> the maze for inactivity, it was boaring.  Given another 3-5 minutes it
> probably would have solved the maze on the first pass (demonstrated later)
> but the rules also state the judges have the option of pulling the plug at
> any time, and this was a case in point.  This was back in the later '80's
> so I doubt we will find out mush about this one.
>
> UCSB had a mouse that used a Polaroid SONAR rangefinder to detect the
> walls.  They actually made it work and it ran fairly well if I remember
> correctly, but the SONAR unit sucked power ina major way and they had a
> very limited run time on the maze.  They could not pack enough power to
run
> for a ful 15 minutes.  Again, technology has advanced since then.
>
> One of the attributes I appriciate about MicroMousers is the learning from
> past efforts.  The design community has been very open about what they
have
> done in the past and there are people like Dave Otten that promote the
> activity by sharing technical information.  This has kept the MicroMouse
> designs progressing, instead of re-inventing the wheel every year.  I hope
> we continue down this path.
>
> I would be opposed to a seperate category for vision based mice, but would
> rather see them compete head to head.  There are other competitions for
> vision systems (vehichle based).  I do think there is pleanty of room for
> this.
>
> My long term goal is to shrink the electronics down to where it will fit
on
> a slot car and run off a couple of lithium batteries.  We could carry our
> maze solving rodents around in our shirt pockets and turm them loose at
> resturants when the service wasn't up to snuff ;-).
>
> John
>
>
>
> At 09:28 AM 6/30/00 +0100, A Johnstone wrote:
> >
> >Piers Plummer and I have discussed several times using a camera this way.
> >The idea was to put a camera on an electric arial and extend it up to
> >where the whole maze could be seen, pull it down again when the image
> >processing has been done and then just run to the centre. However, as
long
> >as the scoring system rewards only the fastest run, there's no real
> >benefit because in ten minutes a mouse with good enough dynamics to make
a
> >fast run also has enough time to exhaustively explore the maze.
> >
> >On a a related note. My mouse EMMA (Adrian's micromouse engine,
> >backwards; also my daughter's name) comprises two pancake steppers, an
> >ADSP 2181 DSP chip and an Omnivision video camera which delivers digital
> >data without any need for an external ADC. The idea is to build a very
> >slim mouse with no extraneous arms hanging off of it so that it will have
> >excellent dynamics and will be able to run up the middle of a diagonal
> >without needing to slew round corners. At the moment, this mouse
comprises
> >a pile of components and I don't even know if the motors will develop
> >useful amounts of torque so I don't claim that this is the last word, but
> >I do hope that in a few years time we'll have a new competition category
> >for vision guided mice. (BTW I did my PhD in machine vision so this isn;t
> >a complete fantasy. All I have to do is find the time to build the
> >thing...)
> >
> >                          Adrian
> >
> >
> >On Fri, 30 Jun 2000, John Billingsley wrote:
> >
> >> Sorry - for decades the rules have said that the mouse must 'negotiate
the
> >> maze - no jumping etc'.
> >>
> >> One proposed gizmo in the early days was to use a high-mounted camera
and
> >> map the maze as a panorama - something the Finnish Microsaurus did by
> sonar.
> >>
> >> Now that video cameras are so small, how about it?
> >>
> >>
> >> Cheers
> >>
> >> John
> >
> >Dr Adrian Johnstone, Senior Lecturer in Computing, Computer Science Dep,
> >Royal Holloway, University of London,  Egham, Surrey, TW20 0EX, England.
> >Email a.johnstone@rhbnc.ac.uk Tel:+44(0)1784 443425 Fax:+44(0)1784 439786
> >
> >
> >
> ******************************************
> * John L. Wright, Jr.
> * Product Apps. Engr. Dept. Mgr.
> * Cypress Semiconductor
> * (408) 943-2886 xjw@cypress.com
> ******************************************
> * IEEE/R6/Central Area Chair 1999-2001
> * IEEE SF Bay Area Council  SCV Director 1998/99
> * 1556 Halford Ave #298
> * Santa Clara,  CA 95051
> * j.wright@ieee.org (email for life)
> * (408) 993-7227                 (personal voicemail)
> ******************************************
>
>
References: