|
[DPRG] digital compass in my car went crazy in Canada
Subject: [DPRG] digital compass in my car went crazy in Canada
From: Chris Jang
christopher.jang at yahoo.com
Date: Wed May 21 14:35:26 CDT 2008
> Don't know if this is related, but I borrowed a Brunton recently
> from the SMU geology department that had just been used for some
> field work in Alaska, and I was surprised to find that the correction
> for magnetic variation pointed almost due east! In Dallas the
> correction is about 4.5 degrees east, so everything was off
> significantly until we reset the compass.
Hi David, I noted a reading bias towards east until well south of the Canadian
border. So I was heading due S but the compass kept reading SE. I wonder how
these devices work in Canada? Maybe they only sell them with a GPS receiver.
> So Chris, we would love to see an outdoor robot do the DPRG
> ORC challenges with vision alone, optical flow, terrain mapping,
> scene recognition or whatever. Lots of builders think that
> ultimately the whole navigation process might be accomplished
> visually. But it's not clear how to get from here to there.
> How are you progressing?
I would love to see the same things too. The last few months have
been between 70 to 90 hour work weeks. I have been swamped with
my job.
Here's what I've learned in the last eight months.
You were right - by that I mean the solution to this outdoor
robot problem. And pretty much the solutions adopted by the DARPA
challenge robots are also correct. They are all designed with
practical consideration of current technology.
Statistical methods are not magic. They are necessary when data
quality is low or when ... theory is inadequate. We just may not
understand the problem. In this case, statistics may be the only
tool we have.
Computer vision has many deep limitations. It splits into two main
areas: data mining and scene registration. The former is about
classification, clustering and search. The latter is estimating
geometry. All of this is immature.
Regarding robots - vision tends to have high computational
complexity. It is far cheaper to use odometry and an IMU instead
of relying on a bigger computer that can see. I think now that
vision does make sense for doing things that are otherwise
impossible without it. So a cruise missile might need to see to
get really precise terminal guidance. Or a robot on land might
need to read street signs or look down a road into the far
distance. But in both cases, vision augments RADAR, LIDAR and
conventional statistical filtering to improve performance. It's
not a substitute for traditional approaches.
More information about the DPRG mailing list
|
