Model calibration / fitting

[Morgan Sangeux]

Hi,

As far as I have understood, two steps are required to calibrate a model for a given subject for inverse kinematics/dynamics. First we need to scale the dimensions of the model to the subject dimensions and then set the model parameters and markers position to the experimental data.
What are the model parameters optimized in this steps?
Is it only the general coordinates which govern the pose estimation or all the characteristics of the joints?

For example the knee joint pose is governed by only one coordinate, the knee flexion. However, the knee joint of a particular subject can have a non-zero static rotation about the X or Y axes. In the case of a genu valgum for example.

Will the model fitting operation try to adjust those parameters as well or not?
If yes, how do we control this (weights)?

Thanks,

[Morgan Sangeux]

I have tried several things and I believe it is not possible.

I may have an idea but it would be easier with an internal OpenSim feature.

This idea would be to have some coordinates set as 'Calibration only'. Only the static pose estimator would be allowed to set those coordinates. Once a model calibrated these would be reversed to constant values.

At the moment I am thinking of having two models linked, one with the dummy coordinates (knee varus, knee rotation etc.) and one with the final dof for inverse dynamics. I would calibrate with the first, copy the dummy coordinates as constant in the second and continue the different steps with the second.

This is a bit cumbersome to do manually or from Matlab. It would be much easier if OpenSim would take care of this.
Feature request?

[Ayman Habib]

Hi Morgan,

Indeed, scaling doesn't change the degrees of freedom, or joint definitions, instead it computes scale factors for existing bodies and applies them. You can do what you want (introduce coordinates) manually for a couple joints but to do it for all joints would be cumbersome. Please file a feature request and we'll see what we can do.

Best regards,
-Ayman

[Morgan Sangeux]

Done, thanks.