Knee bending in the wrong direction

AxonSpark

daylanKifky When I was putting the project boxes together, I made sure that I was buying pure brass screws and nuts. I would take a magnet from elsewhere in the hardware store and test to screws & nuts to make sure there weren't merely plated.

Switching the base and right upper leg K-Ceptors actually resulted in an improvement. The right leg started working normally, and the base kept working. I didn't do anything to the left leg, so it was still behaving oddly. What I don't understand is why did this improvement occur?

I'll have to get back to you on the other tests.

AxonSpark

Hey @daylanKifky,

I had a break in my freelance projects, so I was able to perform the Physics Toolbox tests, and the K-Ceptor vector tests:

With the Physics Toolbox I did 3 tests:

Ceiling to Floor, and then moving the phone to a couple of different places in the vicinity of where I did my full biped test.

Ceiling to Chair, and moving the phone over where my feet would've been placed during calibration

Ceiling to Stool & Chair, and moving the phone over where my feet would've been placed during calibration, and also on the chair I used to step down from the stool after calibration.

Additionally, I did these same tests for each K-Ceptor, but I performed them both with the K-Ceptor matching up to the Blender world axes (as that tends to be the position where the K-Ceptor has the most noise), and pointing South (as my feet would've been during the test). So there are a total of 6 tests per K-Ceptor (Sorry if this is testing overkill).

I tested K-Ceptor 0-2, as that one has been pretty reliable, 0-1, since that one seemed to be causing some problems, and 2-2 & 2-3 since they both were the feet in the last round of tests.

Physics Toolbox Tests:

Ceiling to Floor: 00:11
Ceiling to Chair: 01:26
Ceiling to Stool & Chair: 02:28

K-Ceptor 0-2 (originally acted as the Base, but was then moved to the Right Upper Leg)

Ceiling to Floor (World Axes): 03:51
Ceiling to Floor (South): 04:56
Ceiling to Chair (World Axes): 06:25
Ceiling to Chair (South): 07:20
Ceiling to Stool & Chair (World Axes): 09:14
Ceiling to Stool & Chair (South): 10:21

K-Ceptor 0-1 (originally acted as the Right Upper Leg , but was then moved to the Base)

Ceiling to Floor (World Axes): 11:41
Ceiling to Floor (South): 12:58
Ceiling to Chair (World Axes): 14:10
Ceiling to Chair (South): 14:56
Ceiling to Stool & Chair (World Axes): 15:51
Ceiling to Stool & Chair (South): 16:51

K-Ceptor 2-2 (Right Foot)

Ceiling to Floor (World Axes): 18:03
Ceiling to Floor (South): 18:59
Ceiling to Chair (World Axes): 20:06
Ceiling to Chair (South): 21:03
Ceiling to Stool & Chair (World Axes): 22:14
Ceiling to Stool & Chair (South): 23:22

K-Ceptor 2-3 (Right Foot)

Ceiling to Floor (World Axes): 24:36
Ceiling to Floor (South): 25:30
Ceiling to Chair (World Axes): 26:29
Ceiling to Chair (South): 27:26
Ceiling to Stool & Chair (World Axes): 28:20
Ceiling to Stool & Chair (South): 29:21

AxonSpark

Hey @daylanKifky, just checking in to see if you saw my post. Sorry it took so long to get those tests done.

daylanKifky

Hey!
sorry for not saying anything before, I did see them. Is just that is not easy to find out what's the problem, I would need some time to carefully go though the results you are sending (by the way, thanks for the exhaustive report 😉 ). The whole pandemic situation is also taking time from our regular activities, so I might be a little slower than usual.
Thanks for the patience!

daylanKifky

Hi there,
I still have no clear indication to give you based on the tests you sent me. There's some magnetic perturbation close to the ground, but it doesn't seems enough to cause the knee bending you are experiencing.

We will be releasing a new version of the notochord soon, which will include the new sensor fusion algorithm. We can try with it.

And if you have the opportunity of testing your suit in a different location (outdoors preferably) we can also get some conclusions from it.

AxonSpark

daylanKifky Thanks for the update. I have a freelance gig at the moment, but I'm hoping to test out the new software this weekend. Before I do though I have a couple questions:

1 - Will I need to recalibrate my K-Ceptors?
2 - Is this the correct list of commands for installing the new software?

cd ~/notochord
git checkout develop
git pull
scons -j3 debug=1

Thanks!

daylanKifky

Hi,
This new version tries to recover your old calibration data and use it. This works well most of the time but there are certain situations in which a new calibration is needed.
You can try to use the current calibration, and if you have problems recalibrate.

By the way, the complete version 1.0.0 of the addon is already available at our downloads page. Install it as well in order to see the colors representing magnetic disturbances.
Upon startup you will see most sensors as red, that's because they need to gather a few samples of the current magnetic situation, since they perform a secondary "live" calibration while you use the suit. Just move them around until they become green.

The procedure for updating the notochord is correct 😃

AxonSpark

daylanKifky I tested out the default biped configuration with the current calibration and my results were stranger than usual, so I started re-calibrating my K-Ceptors today. However, I noticed that each time I calibrated a K-Ceptor it says that 0 samples were collected. Is this normal?

daylanKifky

Yes, that line dumps the samples collected with the old method. I just forgot to remove it 😅

A good way to tell a calibration was calculated correctly is look at the matrix. If it's numerically similar to an Identity matrix then it's probably OK. Of course this should be confirmed by testing the sensor with the testcube method.

An identity 3x3 matrix is:

| 1 0 0 |
| 0 1 0 |
| 0 0 1 |

so ones in the diagonal and zeroes for the rest.

As you can see the one from your screenshot is OK

AxonSpark

daylanKifky Here's my latest series of tests using the new software with the sensor fusion algorithm:

Since I was getting some pretty big issues with the arms and legs with the original calibration, I went ahead and re-calibrated the K-Ceptors and did more tests. I've probably mentioned this before, but I do my calibrations on top of a cardboard box on a wooden stool (in order to eliminate possible interference in case there are non visible metal pieces in the stool). Unfortunately, this didn't help.

What's odd to me is that despite doing these armature calibrations/tests in roughly the same place, which bones are registering magnetic disturbances tends to vary between tests, and based on the readings I took in the last tests aren't happening where I would expect them to at times. There also seems to be a fair amount of limb twisting happening in the tests with the new calibration.

I'm also unclear as to why certain bones started to change color when I would for instance, move my arms, as I wasn't getting closer to possible sources of magnetic disturbance.

I attempted to try and test the K-Ceptors individually today, but the add-on kept disconnecting them due to a divide by zero error that kept coming up.

As with my more recent tests this was the K-Ceptor configuration I was using:

Head: 2-1
Back: 1-1
Root: 0-1

L Shoulder: 0-4
L Forearm: 1-5
L Hand: 2-5

R Shoulder: 0-5
R Forearm: 1-4
R Hand: 2-4

L Upper Leg: 0-3
L Lower Leg: 1-3
L Foot: 2-3

R Upper Leg: 0-2
R Lower Leg: 1-2
R Foot: 2-2

Hope some of this info is useful.

daylanKifky

Hi!
It's also unclear to me why you are getting those red bones, particularly those in the upper arms.

I don't remember if we discussed this before, but have you ever tried calibrating and capturing without the enclosures? I remember you were using brass screws as I suggested at the beginning, and that you tested them with a magnet before using them. But since you are getting this systematic errors the only remaining part I can think of as causing troubles are those screws. You can try to use the temporary fixing technique I described here: https://forum.chordata.cc/d/98-build-log/28

For the "division by zero error" bug you can delete the Stats node until we have the time to fix it.

Let me know how it goes

AxonSpark

daylanKifky Hey, sorry it's been awhile. Between corporate work, some Blender VFX work, and volunteering during the lead up to the US election, I've been a bit busy.

I was looking through the post you linked to and had a question: Are you using nano tape to secure the K-Ceptor to the rigid foam? I ask because in researching nano tape, it seems like if I have to make any future mounting changes, trying to remove the tape might damage the PCB. Is there an alternative to this that you would suggest? TY!

daylanKifky

Hi!
We are not using actual nanotape, just regular two sided tape. It can be eventually detached without problems.
As a side note to others reading this: we are discussing quick fixes to test some aspects of the KCeptors, the tape would probably not be the best solution for permanent fixing

AxonSpark

daylanKifky

I re-mounted all of my K-Ceptors on foam using double-sided tape, and then put the foam pieces together with hot glue and attached self-adhesive velcro to the bottoms. Then, I re-calibrated them all, suited up for a full biped test, and...well...nuts:

At least the problems I'm having seem to be more consistent now. The knees are still bending incorrectly, and for some reason the Left Forearm, Right Upper Leg, and Right Hand don't seem to be too happy (nor are the joints on the left leg when I move it around). I'm not sure where to go from here. 😕

NakedRabbit

Hi, @AxonSpark, I'm the one who was using nanotape!

And yes, initially I used it to adhere the Kceptors right to the SKIN, because I was trying to keep everything as low profile as possible. That's when I discovered that any tension or torsion on the Kceptor PCB could cause it to get out of calibration! SO I made 3D printed cases for them. I do stick the Kceptors to these cases with the nanotape. I have not had any issues removing the tape, but I tend to do it slowly so I can wash the tape and reuse it.

And, since I made the cases, I've not had to remove the Kceptors, so it's not been an issue. The cases allow me to access the Kceptor for anything I need to without having to remove them. The current design uses elastic straps, like most people have been using, to hold the Kceptors to the body.

The nanotape has not, in my experience, been difficult to remove from the Kceptor, nor particularly adherent to any aspect of the PCB.

Hope this helps!

daylanKifky

Hi!
yes, as you said the knee problem is more consistent in this video.
The only thing I can think of trying would be to make the test outside, I assume these ones were done in your apartment, right?
It would also help if other users with working suits can confirm this problem is not occurring to them. So far I haven't seen anyone else with this particular issue.
Perhaps you could send us the sensors to test them here

ShiroSaki

in blender 2.9.2

QUATERNION Quaternion (WXYZ), No Gimbal Lock.
with the calib box downloaded from the site, the Chordata_node.cpp should edit the line 449 to

Quaternion q(the_quat.q0, the_quat.q2, -the_quat.q1, the_quat.q3);

I will open a pull-request later

daylanKifky

ShiroSaki
Hi!
Exactly which issue should be solved with this modification?

ShiroSaki

daylanKifky after I modified this, in my blender(2.9.2) all the ceptors worked correctly. before the modify, all the ceptor's x and y are not in the right position. when I lift up my arm, the module in blender will rotate. and when I rotate my arm, the module lift up the arm.

NGWORKS

daylanKifkyAccording to the test, the project does have the situation of X axis and Y axis interchange (that is, the rotation along X axis is settled as rotation along Y axis). I think this situation may be caused by the different properties of the coordinate system (the difference between the Blender coordinate system and the sensor coordinate system).

daylanKifky

ShiroSaki

It sounds strange to me. I still haven't tested it in blender 2.9.2, but it sounds unlikekly that they change the sign of one of the Quaternion components from one release to another.

Could you elaborate why you choose to negate the Y component of the Quaternion and why it was causing troubles before?

I will test it with a suit as soon as possible

ShiroSaki

daylanKifky
The calib box provided in the site https://chordata.cc/wp-content/uploads/2019/04/CalibrationBox.pdf
Seems not the same as https://wiki.chordata.cc/wiki/File:Magnetometer_calibration.gif)
Maybe this difference cause the bug?

ShiroSaki

daylanKifky After talking to @NGWORKS we found out that

is not using the same Coordinate System.
Maybe this is the true reason why the ceptor is bending to the wrong direction