simtk.org

Dear Experts,

I am running a baseball pitching motion simulation and got stuck when trying to proceed static optimization and residual reduction.

For the Static Optimization part, I made my own xml files of the actuators and the task for my model, which was modified from Hamner's full body running model.https://simtk-confluence.stanford.edu:8 ... ning+Model

The problem is that the description "Optimization Failed" always appear in the message box although the OS procedure is completed. According to OpenSim's guideline, it means that model actuators may not able to generate sufficient force/torques to match the motion. Although I checked the xml to make sure every coordinate is driven by an actuator (capable of generating very large force/torque), the message box still display OPTIMIZATION FAILED in each time step.

The second problem is that my model suddenly disappears after the first time step when I proceed RRA. The procedure can still finish but all the COM and segment mass results become "-nan(ind)" and residual become :
* FX=-1e+09 FY=-1e+07 FZ=-1e+07
* MX=-1e+11 MY=-1e+11 MZ=-1e+06

Please let me know if you had encountered and solved similar problems. Your kind help will greatly appreciated.

Hi Amber,

I think you should try to solve the RRA problem first. Please, try increasing the tracking gains that were provided to the tool, so that the model does not fall. Also, before doing RRA it is advised to run inverse dynamics and check the residuals at the pelvis. If the pelvis forces and moments are high, this probably means that you are not applying the ground reaction forces properly. Please not that a good RRA run will not require large modifications to your model and motion. If you are successful with ID and RRA, then you should try to solve the SO. Typically, if you do not have large pelvis residuals then SO should not be an issue.

Regards,
Dimitar

Hi Dimitar,

I deeply appreciate your advice. I've checked the tracking gains that you mentioned and modified them from 20 to 1000(pelvis tx,ty,tz). Sadly , the model still disappeared after the first time frame. I also found that my model was not "falling" while performing RRA. I checked the example "leg6dof9musc" which displays model falling in the first procedure, and I could see the pelvis ty coordinate gradually approached zero frame by frame.
But in my case, all of model's coordinates became zero just after the first time step. Is there ant other modification that can fix this problem?

Also, I am not sure if you are aware of this. You should narrow* the time (begin and end) of the RRA analysis in the region where your subject touches the ground. Maybe you are starting RRA based on the IK information, however, in the begging ground reaction forces may not be applied. Then, it makes sense if your model falls, because there is nothing to compensate for the gravity. On the other hand, if the tracking gains are higher then your reserve actuators should be able to balance the model even though it is not physiological (e.g., superman). Please check:

- The IK result
- The ground reaction forces
- Tracking gains
- Reserve actuators (e.g., they might not be strong enough)

However, the first step is to do the inverse dynamics properly and check the residuals at the pelvis. This will tell you if there is something wrong with the fist two (above).

* For the begging time you should consider the first step on the ground plate after the trailing foot is being lifted. Also,for the end time you should choose just before the swing foot touches the ground floor (not ground plate).

mitkof6 wrote: ↑

Tue Oct 13, 2020 2:03 am

Also, I am not sure if you are aware of this. You should narrow* the time (begin and end) of the RRA analysis in the region where your subject touches the ground.

What if the model is standing on the force plates through the whole task and the task is longer than 2 seconds?