CMC tool with slow target does not work properly

[Sietse Achterop]

Hello List,

I created a number of models and created trajectories for it. There are no muscles involved, but a few constraints.
I use the CMC-tool in the slow target mode. For simple models it works fine, you just have to set the weights correctly.

But a more complete model (26 bodies, 29 coordinates/actuators, 3 weldconstraints, 1 contact force with ground) the simulation goes wild.
Here a picture of the model

the model

bb.png (43.17 KiB) Viewed 236 times

I tried to set the weights for CMC as good as possibe, now all weights are at 1000. Also played with the PD parameters, but that didn't help too much.
Tried both 50Hz en 100Hz sample rate in the trajectory.trc file. Making the motion slower helped a bit but not much.

The model and all data can we found in github: https://github.com/SietseAchterop/Rowin ... r/BootBaan.
To test the system use Bootbaan.osim (generated from bootbaan.py) and trajectory.mot (generated from trajectory.py and the IK tool).

On github you can find both a video of the result of the IK-tool and the result of the CMC-tool that shows the error: bootbaan_IK_video.webm and bootbaan_CMC_video.webm.

behaviour of one elbow

bootbaan-cmc2.png (93.21 KiB) Viewed 236 times

This plot shows how the controller fails.
se_inout and se_updown (red and blue) give the intended behavior of the left elbow, and the other two what happens to the elbow when CMC is at work.
In this example the tracking is correct for about 2 seconds and then suddenly deviates violently.

I really don't know what I can do to make the system stable. Is this system too complicated for CMC when using slow target mode?
Note that I don't use muscles (yet) and maybe there are implicit assumptions that there should be some?
I sometimes get this idea when reading the documentation.

Help would be most welcome!

Sietse

[Sietse Achterop]

Simplified the model further, to 11 coordinates. Also 2 of the 3 kinematics chain loops are removed.

I now have 2 models that only differ in one body and one joint.

Simpified model

bootbaan_simplified.png (55.72 KiB) Viewed 175 times

This is the model where the error occurs, which starts with a wild spinning of the upper arm.
When the, single, upper arm and joint is removed, everything is stable.

The models are generated from bootbaan_OK.py and bootbaan_ERROR.py.
The matching CMC configurations are bb_CMC_Tasks_OK.xml and bb_CMC_ERROR.xml.

The trajectory is the same as before, and can be seen as correct both visually and in the data.
ResultsCMC contains the data of trying the ERROR versions.

Maybe someone can verify that the error actually occurs? Here it does on Windows and Linux.