|
[DPRG] Borenstein squares
Subject: [DPRG] Borenstein squares
From: Randy M. Dumse
rmd at newmicros.com
Date: Tue May 20 15:12:12 CDT 2008
Dpa said: Tuesday, May 20, 2008 12:43 PM
> Randy, the Borenstein squares do not require precision 90
> degree turns,
Then it really isn't a square is it?
I rather wonder if Borenstein didn't mean "square" for his
square.
And yes, I am not niave enough to imagine steering in the
straights aren't required as well.
> but rather accurate tracking of the robot's
> position thoughout the run.
What we've been finding on our new platform is that small turns
odometry works reasonably well, but on large turns, like 90
degree turns, we get different numbers each time we turn.
First, being the only two wheeled differential drive robot, one
might suppose we are more likely to turn more accurately than
most. So something is wrong. Seems to me there are several
explanations for this.
Could be our "tractor" like tires are especially poor for
odometry. Our results do see very surface specific. As I
mentioned earlier, our turns suggested encoder differences of
14,000 to 19,000. That's about an error of 5000 counts out of
15000. It is very hard to imagine successful odometry with such
variation on knowing how well your turn did.
Could be our drive train is slipping. Maybe. Not likely.
Could be the omni wheel is dragging and giving bad interference.
Quite possible, but not sure how to isolate.
So assuming our problem with counts is design endemic, meaning
the tires we have, then doing a 90 degree turn must be done with
gyro, compas, or IMU of sorts, because the current odometry
fails here for us.
Randy
More information about the DPRG mailing list
|
