DPRG mailing list archives: [DPRG] Robot contest at the Intl Space Development Conference

Mon Apr 30 17:57:26 CDT 2007

Hello Steve, Everyone,

My robot can run outdoors. Here is a photo of Stella.

http://golem5.org/embedcv/images/stella20070220.jpg

She has no GPS, compass, odometry or bump sensors. She relies
exclusively on a monocular camera (USB webcam) for environment
sensing. The computer vision is feature based from the luminance
image.

As you know (as you have seen me at the DPRG warehouse), this
work is immature. You saw Stella really working for the first
time last Tuesday (and that was not much). I think I can get
things together well enough for Stella to do something by the
26th. But what follows is probably just about impossible to get
together by then (the wish list). I find a lot of testing and
tuning is required to have this stuff work. But...I will try.

Right now, there is simple obstacle avoidance. Stella looks for
a drivable corridor clear of visual features from the bottom of
the image frame. I've also done simple convolutions to detect
soda cans. So if there are visual waypoints like traffic markers
or flags, then it is possible to design a filter to detect these
and drive towards them (have not done this yet).

However, the turnaround on software and especially filter design
is pretty long in my experience. Unless there is test imagery to
work with early, it really isn't possible to design for it. So
my bias is towards making Stella as "tactical" as possible. The
tactical ability is obstacle avoidance and has the highest
development priority.

If you can take some photos of the site, that would be very
helpful. Stuff I look for: uniformity of the ground surface,
both physically and in luma/chroma; painted lines on the ground;
boundaries of contest area; tall objects like poles and buildings
in and around the contest area which might cast shadows.

Ideally, we'd have a completely enclosed contest area like a
large open maze of some kind. Then robots with any of bump
switches, sonar, IR distance sensors, or computer vision could
equally detect the walls. And there would be complete safety as
robots could not run off. But then it is logistically difficult
and expensive to assemble a contest course like that on-site. So
we will probably have something more open (which is really more
challenging and interesting anyway).

Chris

-----Original Message-----
>From: "R. Steven Rainwater" <srainwater at ncc.com>
>Sent: Apr 30, 2007 5:54 PM
>To: DPRG <dprglist at dprg.org>
>Subject: [DPRG] Robot contest at the Intl Space Development Conference
>
>The ISDC folks have confirmed that we can use the vacant land adjacent
>to the InterContinental Hotel for an outdoor robot contest. What I need
>to do now is find someone with a GPS who's willing to visit the site
>with me so we can pick out some likely waypoints for a contest. If
>anyone is able to do that tomorrow afternoon before the RBNO email me
>and we can work out a time and place to meet.
>
>The contest will be on Saturday, May 26. The contest format will be
>similar to the outdoor event we did at the Science Place last year
>(though without the rain hopefully). If you have a robot that might
>participate, how about posting something to the list about its
>capabilities. I'd like to make sure the contest rules allow the widest
>participation possible.
>
>-Steve
>
>_______________________________________________
>DPRGlist mailing list
>DPRGlist at dprg.org
>http://list.dprg.org/mailman/listinfo/dprglist