Calibration Questions

daylanKifky

Hi,
Here's two more tests to better understand what's happening inside the K-Ceptors.

Please start by compiling the latest version of the notochord with

git pull
git checkout develop
scons -j3 debug 1

Pick just one KC, we will use it as our test subject until we figure out what's happening.
You will have to perform the movements like in the magnetometer procedure, and record the results. Do these tests on the same spot you used for the last calibration.

For the first test run the notochord with this command:

bin/notochord --scan -t file -f kc_dump.csv -v1 -r

Explained:

--scan will search for connected KCs instead of looking into the configuration file
-t file put the transmision data on a file
-f kc_dump_calib.log use a file named like this
-v1 verbosity level 1 (include debug statements)
-r transmit the raw data, without sensor fusion (sidenote: this flag was used to perform the calibration on the previous version, not any more)

For the second test a slightly different command:

bin/notochord --scan -t file -f kc_dump_nocalib.csv -v1 -r -k1

This time we are using the -k1 argument to tell the notochord to use the K-Ceptors as in Revision 1 . This is just a hack to make it not find the EEPROM, thus logging the uncalibrated data

Please post both files. You will find them in bin/logs. And also include the captures of the console output of the notochord for both cases.

AxonSpark

Ok, here's what I have so far:

Simple Test

Magnetic north and true north do roughly match, although it looks like magnetic north deviates by about 10 degrees from true north.

Complete tests:

Here's the Dropbox link to the CSV results I got from the Physics Toolbox Sensor Suite:
https://www.dropbox.com/sh/ckdmg6d2d6rhzif/AADiviguKA-PuL0cvAjkRVAua?dl=0

K-Ceptor Test Issues:

Unfortunately, I wasn't able to complete the K-Ceptor tests. I was able to pull/checkout the latest version of notochord, but when I tried to run the scons -j3 debug 1 command, this happened:

I also tried running it in the main directory, and had this result (note: the same message comes up if I'm running it in notochord/bin):

I'm not sure what to do about this.

I decided to try and run the first test anyway, and I got this result back and then it aborted (note if I ran the command without ./notochord it didn't recognize the scan command):

Let me know how I can fix this.

daylanKifky

AxonSpark

😅 My mistake, I forgot an = sign. The command to compile should be

scons -j3 debug=1

AxonSpark

daylanKifky Ok, so I've completed all the tests now. The CSV's for everything can be found here:
https://www.dropbox.com/sh/ckdmg6d2d6rhzif/AADiviguKA-PuL0cvAjkRVAua?dl=0

First Notochord Test:

The first time I ran the test I got a huge list of those debug messages about "Read wait", but subsequent attempts didn't result in that many.

Second Notochord Test:

Unfortunately, re-running the second test didn't reduce the amount of debug lines that came through on the terminal, and there are a fair amount of entries in the CSV file for this test that have this debug data.

daylanKifky

Thanks for taking the time to perform all these tests!! 🙌 🙌
Although some of the datasets give me completely unexpected results. I suspect that I wasn't clear enough when describing the movements you had to do.

Smartphone test.

The first one is OK and, as expected, the magnetic field is not fluctuating.

The second dataset is slightly different than what I expected. The results of this test should show 4 transitions, like in the picture, but yours has many more.

Please double check you are exporting the Magnetometer data, and you are making a procedure like in this video.

K-Ceptor dumps.

Here I expected to get an spheroid from the data, like in the images generated during calibration. What I get instead are two intersecting semicircles.
Here you can see the 3D representation of the the nocalib dataset, the other one looks similar

Are you making the same movements you used for the magnetometer part of the calibration? as in this gif:

PS: don't worry about the debug data on the logs, it can be easily filtered.

AxonSpark

daylanKifky It looks like I misunderstood your instructions.

Put the phone on a flat surface, and make four 90° turns, ending on the same direction you started. Between each turn hold it still for some seconds. This will help us determinate if the field has a significant vertical component, in other words, if it's pointing upwars

At the time I did the tests (both for the smartphone and the K-Ceptor), I thought that I needed to start with the phone/K-Ceptor on the flat surface and then then rotate it 90-degrees on one side so it was perpendicular to the flat surface, and then return it to it's original position, and then repeat the process for the remaining 3 sides. However, upon reading it again I think I see where I went wrong.

So for these tests, am I supposed to 1) perform the same full procedure that is used when calibrating the K-Ceptors, or 2) keep the phone/K-Ceptor parallel to the flat surface and rotate along the yaw axis 90 degrees and repeat until I come back to the side I started on?

daylanKifky

AxonSpark 1) perform the same full procedure that is used when calibrating the K-Ceptors

This is the procedure to be used with the notochord tests

AxonSpark 2) keep the phone/K-Ceptor parallel to the flat surface and rotate along the yaw axis 90 degrees and repeat until I come back to the side I started on?

For the test with the physics toolbox, using your smartphone.

AxonSpark

daylanKifky Ok, I've redone the tests. Here are the CSV's:
https://www.dropbox.com/sh/ckdmg6d2d6rhzif/AADiviguKA-PuL0cvAjkRVAua?dl=0

First Notochord Test:

Second Notochord Test:

daylanKifky

Hey,
thanks for the new readings and for your patience through all this process!
I now have the data I need. I might need a couple of days to figure out what's going wrong

AxonSpark

Ok cool. Hopefully it's an easy fix. I can't wait to test out the full biped armature! 😁

daylanKifky

AxonSpark
I think I found the problem!
Or at least I found an issue with the way the accelerometer and gyro calibration data is applied. Hopefully that's the reason of your calibration problems.
I'm working on a fix, will let you know as soon as it's ready

daylanKifky

At the moment the fix is partially implemented at the branch b_bad_calib.
I should be able to complete it on the following days. If you want to test it now you might already get a better result.

this will checkout the correct branch and compile:

git pull
git checkout b_bad_calib
scons -j3 debug=1

(you will need to re-calibrate the sensor)

AxonSpark

daylanKifky I compiled the new branch and performed the calibration, but now the plugin in Blender isn't able to receive data. I seem to be connecting without a problem, but when I click the "Receive" button nothing happens.

daylanKifky

I've just pushed the complete fix. Please pull the repo and recompile, and tell me how it goes.
(you will need to re-calibrate the sensor.......... again 😁 )

If you still find problems please send me a new set of CSV's like the one you did before.

AxonSpark when I click the "Receive" button nothing happens.

Is this happening when you run the notochord with the --scan flag?
If that's the case then is normal: the current add-on is not really smart, it just assigns the rotations it receives when the label of the object matches the label of the K-Ceptor.
When using --scan the notochord assigns labels based on the position of the KC on the hierarchy, something like kc-CH_2-1 (kceptor on CHannel 2, with value 1).
The object on the testcube.blend is called Unico. In order to use it with --scan you should change it's name.

daylanKifky

While calibrating a set of sensors I realized that there's a little problem with this last fix.

If you plug the KC and proceed right away with the calibration you will probably get wrong results.

I'll solve it on the next commit. In the meantime a temporary solution is to make a dry run of the notochord in normal mode before calibrating

AxonSpark

daylanKifky Good thing I waited to calibrate! 😃Ok, so I pulled/compiled the latest version, did the notochord dry run, and then calibrated the K-Ceptor (and used the proper K-Ceptor name in the testcube file).

The results I saw are looking REALLY good! 😁When I turn and hold the K-Ceptor, I'm not seeing any of the large constant drifts that I had before. In fact, I'd say it's pretty solid, as not only does the test object stay still, but there also seems to be a significant reduction in the noise coming through as well (if it weren't for the minuscule changes in rotation values in the object panel, I wouldn't have known there was any noise at all).

The only visibly noticeable noise I saw was when the K-Ceptor was in it's resting position, but even so it doesn't seem to be that significant.

Thanks for your speedy updates!

daylanKifky

AxonSpark The results I saw are looking REALLY good! 😁When I turn and hold the K-Ceptor, I'm not seeing any of the large constant drifts that I had before.

That's great news!!
Thanks to you for the detailed feedback!, I wouldn't be able to even find this bug without the data you provided.
Let's celebrate by creating a new [Solved] tag for this forum and assigning it to this thread! 🎉🎉🎉

Once I finish solving the little issue that popped up with this modifications I'll create a new release of the notochord.
And will also update the information on the calibration section of the wiki, and add the procedure to the capturing section.

daylanKifky

I was finally able to update the information on the wiki (capturing and calibration sections). Let me know how it goes with the full body capture!

AxonSpark

daylanKifky My freelance gig started up again, so I'm going to recalibrate the remaining sensors this weekend. I've also ordered the battery pack and plan on going to the local REI to get the straps/harness I need (I figure if any one would have that sort of stuff it would be a camping supply store). Still need to order some sheets of anti-static foam.

One thing I noticed in the photos is that the K-Ceptor's are secured with screws. Do metal screws not interfere with the sensors after they've been calibrated? Or should I look into some plastic screws? Also, is it safe to use the TP holes on the Hub for screw mounting to a project box?

daylanKifky

AxonSpark Do metal screws not interfere with the sensors after they've been calibrated?

If the screws are fixed to the KC they can be made out of metal, but you will have to calibrate after you attach them together.

No to say that non-ferrous metal interfere very little with the magnetic field. Some screws contain little, or no iron.

« Previous Page Next Page »