DPRG mailing list archives: [DPRG] Further Quandries on GPS...

Mon Jan 28 17:05:53 CST 2008

Howdy,

Kenneth Maxon wrote:
> David & DPRG,
>
> There is to much rounding error applied to the numbers in the first link in
> your e-mail below to tell if these errors are on the order of what I am
> seeing or not.  Any chance you could shoot me a set of coordinates defining
> the outer boundaries...?
>   

Well, I couldn't find the actual data for that plot so I made another.
I setup the GARMIN Etrex GPS with my laptop out in the SMU
Geosciences commons this afternoon and collected data for
40 minutes.   I converted the data to decimal degrees and plotted
the locations.  The plot is here:

<http://geology.heroy.smu.edu/~dpa-www/gif/gps_garmin_28jan08_dec_3b.png>

The whole range of variation was about 8 meters from min to max,
which more or less corresponds to my experience with the robot.  

This is not a WAAS enabled GPS like the new Garmins we have here
in the department.  I'll get one of those and try the same experiment,
but I can only assume that the drift will be smaller than the plot above,
else what's the point?

I have not had great luck navigating my outdoor robot with GPS alone. 
The instantaneous GPS error is often larger than the spaces through
which the robot is trying to maneuver, i.e., 8 foot error trying to get
through a 6 foot opening.

So for navigating in small spaces, i.e. under a square mile, like might
be found in a DPRG Long Haul or SRS Robomagellan contest, jBot
translates all the GPS lat/lon coordinates into robot X,Y coordinates
and navigates with wheel odometry and its IMU, and only uses the
GPS to get the local Magnetic Variation, which is needed so that  the
navigation targets match map lat/lon coordinates.

I'm hopeful that !!Dean and Chris will get a Kalman Filter working which
includes GPS that they will share with the dprglist, as that clearly seems
like the way to go.

Did you ground the GPS ground plane to the robot itself, or just to the
GPS?  I had to ground the aluminium channel that the GPS is sitting in
to the main supply, and also mount it as high above the chassis as
possible without interfering with the IMU. 

Might be interesting to put a handheld GPS next to the robot and dump
its output to a laptop and see'f you get the same drift for both. 

cheers,
dpa

> I have finally gotten back to running another set of experiments and I am
> still receiving disappointing results.
>
> In the ground plane overkill dept, I added a 18" long by 8" wide by 3/16"
> thick solid aluminum plate to the top of the robot's ~4' mast.  I terminated
> the ground return lead of the GPS module (SiRF III chipset) to the ground
> plane using a <4 milliohm* connection.  *Note: Measured with a 4wire setup.
> (4 milliohms is no where near perfect but should suffice for this test).
> The GPS module w/integrated antenna I set right down onto the ground plane
> separated by only ~1/8" acrylic.  I powered up the entire system and left it
> sitting for over an hour and then reset only my processor system, leaving
> the power to the GPS system un-interrupted.
>
> I allowed the system to gather and plot data for ~3hrs this afternoon.   As
> always I verified a couple of random samples through out the data capture
> period against Google Earth to make sure everything was in range with my
> processing routines.
>
> During data capture the least number of satellites in view was 12.  My
> software tracks how many of them have sufficient signal strength and tracks
> a min and max number of satellites used in the position calculation.
> During the 3hr time span, the number of active satellites never dropped
> below 6 and typically remained at 9 or 10 used in the active calculation.
>
> Interesting observation:  During the time periods of most significant error
> there were never less than 9 satellites in use in the active calculation.
> During the entire run the HDOP never dropped below 2.4m and there is no
> significant correlation between the HDOP or PDOP values and the error form
> true position.   I have a plot on the display of my robot but don't have a
> way to get that plot back into a PC currently, I'll attempt to forward that.
>
> During the ~3hr run similar errors were observed as those previously
> reported.  Today typical values remained in a 47ft by 23ft bounding box,
> with several forays of error paths when plotted out over 96ft from the true
> actual position**.   **Note: the "true actual position" that I am using
> comes from running an averaging function for 8hrs continuous and taking that
> value and pumping it into Google Earth for a verification.   It is more than
> accurate enough to determine error at the level of which I am seeing them.
>
> Sooo...   This next round has been somewhat disheartening with the addition
> of the ground plane and the observations of lack of correlation from
> reported PDOP and HDOP to error functions.   Also the lack of correlation of
> number of satellites in view to error in signal is also quite disheartening.
> (I should qualify that last statement as there is a very strong correlation
> there that appears when less than 6 satellites are used.)
>
> Sigh...   Just thought I'd throw out some additional data for anyone
> interested, or anyone with further ideas...
>
> I will be back in a much further South Latitude on business here in a few
> weeks and will attempt to find a way to get a subset of CCAs through airport
> security to test much closer to the equatorially based WAAS satellites that
> we can't seem to receive "here in the middle" with either GPS that I have.
>
> -Kenneth
>
> -----Original Message-----
> From: dprglist-bounces at dprg.org [mailto:dprglist-bounces at dprg.org]On
> Behalf Of dpa
> Sent: Wednesday, January 02, 2008 9:55 AM
> To: dprglist at dprg.org
> Subject: [DPRG] Further Quandries on GPS...
>
>
> Howdy,
>
>
> Here's a plot of 40 minutes worth of wandering GPS locations recorded with
> a handheld Garmin Etrex sitting in a fixed location:
>
> <http://www.geology.smu.edu/~dpa-www//gps/gps_40min.png>
>
> I need to convert the lat/lons to something that makes more sense
> (i.e. feet or meters) but you can see that it doesn't drift anything
> like what Kenneth is getting.
>
> Kip wrote:
>   
>> The first thing I would try is a bigger ground plane under the patch
>> antenna. You should have a minimum of 40mm x 40mm, but anything bigger
>> is better and it needs to be grounded to vehicle ground. Bigger should
>> help reduce ground reflections.
>>     
>
> I found I had to do this on jBot to get reasonable GPS data, and also
> mount the GPS rather high above the rest of the robot platform.  Here's
> a picture:
>
> <http://geology.heroy.smu.edu/~dpa-www/robo/Encoder/imu_odo/jbot_imu_odo_02.
> jpg>
>
> that shows the GPS (yellow and black device) mounted in a grounded aluminium
> channel which shields it on 3 sides, which is what I eventually had to do
> to get a reasonable performance.
>
> FYI, jBot continously calculates its position from odometry as well as GPS
> and
> then calculates the difference between the two locations.  In a typical 15
> minute
> run the two never drift more that 96 inches from each other, and are
> typically
> in the 17 to 54 inch range.  (I think 17 inches is the quanta of the
> smallest
> measureable GPS increment with the handheld garmin).
>
> HNY,
> dpa
>
>
> _______________________________________________
> DPRGlist mailing list
> DPRGlist at dprg.org
> http://list.dprg.org/mailman/listinfo/dprglist
>
>
>   