DPRG List  

[DPRG] Meeting re-cap

Subject: [DPRG] Meeting re-cap
From: David P. Anderson dpa at io.isem.smu.edu
Date: Wed May 10 16:13:17 CDT 2000


Clay writes:

> At the end of the meeting Clay Timmons
> re-programmed his robot from the last contest
> to run a 4 meter (157 inch) square course.  This is part of a
> test/calibration procedure developed by the University
> of Michigan called UMBmark.  A drain in the floor
> made for a less than level surface.   The course
> was setup to run around the drain to avoid the dip in
> the floor.  Clay taped a large piece of graph paper to the
> floor to record the test results.   The X and Y
> positions for each of 5 clockwise runs were recorded.
> (all values in inches to the nearest .1 inch)
>                    X           Y
>   run 1    +0.4       +6.0
>   run 2    -1.2       +8.9
>   run 3    -2.8     +10.6
>   run 4    -3.7       +9.3
>   run 5    -1.0       +4.3
> Testing was not very rigorous but it will
> serve as an intial benchmark to compare
> with future improvments.

Cool!  I especially like the graph paper chart of the results!

Your robot has a UMB "center of gravity" error of these five 
clockwise runs of 

	X error = (0.4-1.2-2.8-3.7-1.0)/5 =  1.66
	Y error = (6.0+8.9+10.6+9.3+4.3)/5 = 7.82

The next step is to run the same course counter-clockwise five
times and collect the same data.  Going around the square both
clockwise and counter-clockwise allows wheel base errors to
be separated from wheel size errors, and this is the really 
nifty thing about this benchmark.

If the errors for the clockwise and counter-clockwise runs are
symmetrical (i.e., X becomes -X and/or Y becomes -Y) then the
calibration error is in the definition of the robot's wheel-
base, the distance between the two drive wheels.  This number
should be tweaked.  Increasing it will cause the robot to
revolve slightly further, decreasing it will cause it to revolve
slightly less.

If the errors for the clockwise and counter-clockwise runs are
not symmetrical (X and Y remain the same sign) then the calibration
error is caused by a slight difference in size between the wheels,
and those numbers should be adjusted.  Borenstein suggests that
a small constant be added to the size of one wheel and subtracted
>from the other.

Ideally the errors for the five runs should scatter evenly around
X,Y = 0,0.  In this case, the calibrations are as accurate as they
are likely to get, and the variation in error is caused by random
bumps and irregularities in the path across the floor.  This is sort 
of like firing a group of shots at a target to see the spread of the
pattern when sighting-in a rifle scope.

Here are the errors from my robot SR04 when it was last calibrated
last summer:



SR04 Calibration 06 Jul 99
clockwise:         X error     Y error

                   1.0         -1.5
                  -2.5          1.75
                   2.0         -1.5
                  -0.75        -0.75
Average error      -.05        -.5       (UMB "center of gravity")

counterclockwise:  X error     Y error

                   -1.5        -2.0
                   -2.5        -1.0
                   -2.0        -1.5
                   -2.5         0.0
                   -3.5         0.0
Average error      -2.5        -0.9      (UMB "center of gravity")

More information about the DPRG mailing list