Inverse Dynamics: Muscle forces without GRF?

[Michael Mauersberger]

Dear helpers,

at the moment I am working on a project for my diploma thesis concerning the interaction of humans to structure. Therefore I need Ground Reaction Forces as a result of a human's movement, e.g. walking. That means the input should be the movement and the output the GRFs. As I know there has to be recorded the GRF, I wondered what results I would get if I just turned off the external forces within the tool "Inverse Dynamics" or "Analyze" (I hope it's the same) in OpenSim (3.2 and 3.3).

So I tried out with Tutorial 3 and got the following muscle forces (via Force Reporter):

The uppermost are the "ercspn" muscles (spine muscles) and hold the largest force while the lower muscles have lower forces. It seems like the whole body is "hanging" at the trunk so that he is not falling due to the gravity.

So here are my questions:
- How does the "Analyze" oder "Inverse Dynamics" tool ensure that the model is not falling due to gravity without external forces? Is there an additional force keeping the model close to the markers? Or occurs the "hanging" effect because of the parent children tree of bodies in the model? (spine is direct child of ground)
- Is there a way to apply some residual actuators as discussed in topic viewtopicPhpbb.php?f=91&t=5956 to account for the missing GRFs?

Though the GRFs can't be validated so well this way it would help me to estimate them without measuring.

Thank you so much for your efforts!

Regards, Michael

[jimmy d]

Inverse Dynamics computes generalized coordinate forces that generate the input motion. If you don't include external forces, the residual (error) forces between and the pelvis and the ground account for the effect of not falling. All other coordinate forces will be as if the limbs are swinging in mid air. Fine if you are analyzing swing, but useless if you need to run analysis when the foot is in contact with the ground.

If you ran a forward simulation — which takes controls/forces and generates movement— your model would just fall. No external forces are stopping it from falling.

If you want to try and predict the GRF, you will have to add Contact elements to the feet and make the model work on a correspond contact plane. This would be very difficult. A single contact element would not capture the 'true' external load, so how you define your contact would be difficult. The anybody group did a webinar on this, and you should check it out.

Cheers.

[Aaron Godfrey]

Hi James,
I'm trying to do something similar to what you allude to in your response, but I'm a bit lost on how exactly to do it. Can you elaborate on how you would predict GRFs? My model isn't defined by muscles or actuators or anything: just the .mot acquired from motion tracking. I suspect that this makes things more difficult, but difficult is okay as long as I know there's a viable path to a solution.

Right now I'm trying to predict the effective GRF you'd find from projecting the pelvis onto the platform. The method I've been pursuing involves using a prescribed motion to enable forward dynamics for a specific .mot, but I'm not so sure forward dynamics is the answer anymore.

Thanks!

[Ross Miller]

Hi Aaron,

A lot of studies have tried to do what you described (estimate GRF from kinematics when GRF data are absent or incomplete). It's a very difficult problem, especially for walking with the double-support phase, but any data-driven analysis of a model with branching chains will be tricky. Some examples:

https://www.ncbi.nlm.nih.gov/pubmed/1769975
https://www.ncbi.nlm.nih.gov/pubmed/17943488
https://www.ncbi.nlm.nih.gov/pubmed/18672243
https://www.ncbi.nlm.nih.gov/pubmed/24835471
https://www.ncbi.nlm.nih.gov/pubmed/26979885

I haven't tried this in a while, but you should be able to define contact elements on the foot in your .osim file and ask OpenSim to output the contact element kinematics from the generalized coordinates of the motion. You could then calculate contact forces as functions of those output contact element kinematics (e.g. Fz = some viscoelastic function, Fx and Fy = some model of friction), then the GRF are just the vector sum of those forces.

The GRF will depend very strongly on how the foot was modeled in your input data, how many elements you define, and what the parameters of those elements are (e.g. stiffness, damping, friction). Hatze (2002) talks about why this is hard:

https://www.ncbi.nlm.nih.gov/pubmed/11747889

Hope this helps,
Ross

[Aaron Godfrey]

Hi Ross,
Thanks for the quick response and the references!

The problem I am attempting to solve is likely a much simpler version of the one you hint at. First, I'm examining exercises: so there's no walking, just a subject standing and performing various workouts, like squats and deadlifts.
Second, I'm primarily concerned with what the GRFs would be if the pelvis were projected onto the platform. In this example, all forces are acting through the pelvis, not the feet, so I suspect that would greatly simplify the problem (one or two contactspheres vs. 12 on each foot, as in one of the papers you linked). Being able to calculate (from a .mot) an analogy to the GRF that acts through a center of mass would likely be an extremely similar problem that, I suspect, would translate fairly well into what I am trying to do.

If you've got further insight into that, I'd appreciate your sharing! Thanks!
-Aaron

[Ross Miller]

Hi Aaron,

Sorry for my ignorance, but by projecting the pelvis onto the platform, do you mean you want OpenSim to apply forces and moments on the pelvis, such that the model moves according to the input kinematic data? I think that could be done by running IK then ID, but I'm not sure the output from ID in that case would be meaningful (I think the pelvis would have very large residual forces).

GRF can be calculated from the body segment masses and accelerations as:

F = m1*(a1 - g) + m2*(a2 - g) + ... + mN*(aN - g)

where F is the GRF vector, mi and ai are the mass and acceleration of segment i, and g is gravitational acceleration (alternatively, you can probably ask OpenSim to just output the acceleration of the center of mass). This is what Bobbert et al. (1991) did for running; some of the results are quite good depending on your standards for "Good" =). The results will of course depend heavily on data quality/processing/modeling/etc. However, to do inverse dynamics afterward then I think you'd need to also calculate or assume the location of the center of pressure.

Hope this helps,
Ross

[Simon Jeng]

Hi Ross,

I am predicting GRF and GRM during squat lifting. The feet are static during lifting, so I built a weld constraint between each foot and the ground and the weld point is not at the CoP because I do not know where is CoP. I successfully got a set of forces by inputting the motion data to 'Analyze' tool. However, I know that the result is not correct because I wonder the GRF and GRM at CoP. Do you know how to get the position of CoP and corresponding GRF and GRM in my case?

Thanks!

Regards, Simon

[Adrián Schmedling]

I am predicting GRF and GRM during squat lifting. The feet are static during lifting, so I built a weld constraint between each foot and the ground and the weld point is not at the CoP because I do not know where is CoP. I successfully got a set of forces by inputting the motion data to 'Analyze' tool. However, I know that the result is not correct because I wonder the GRF and GRM at CoP. Do you know how to get the position of CoP and corresponding GRF and GRM in my case?

Hi Simon,

I am working on similar research. At this point I've done the same steps as you (1-Create a model with WeldConstraints on both feet with ground, 2-Use a .mot file as input and use 'Analyze' tool to obtain the forces and moments generated in the WeldConstraints). However, I know the result is not correct because the sum of the vertical forces should be near the weight of the model and is not the case. Also I think this is not a problem of CoP position. Do you found any solution to this problem?

I found this article https://www.sciencedirect.com/science/a ... 9015005278 which explains a method to obtain the CoP position and force, but I am searching for a easier solution

Regards, Adrian