high speed tracking errors

[Ryan Bakker]

Hello everyone,

I am simulating a single leg jump landing using CMC. I am interested in extracting the quadriceps, hamstring and gastrocnemius muscle forces.

I have ran scaling, IK, ID, RRA but am struggling with CMC, specifically the Perr and reserve actuator for the ankle. This makes sense because GRF is large (~4x body weight) and the high speed at which the angle angle changes when it hits the ground (~60 degrees in <100ms).
I have locket MTP, and Subtalar joints and both kinematics and GRF are filtered at 15hz.

The general guidelines are: joint angles are "OK" if they are tracked within 5 degrees and "good" if within 2 degrees. On all my subjects the Perr term for the right ankle (the landing ankle)
shoots above the "ok" range sometimes as much as 9-10 degrees. A closer inspection of this error is that the large Perr is caused by the large rate of change in the ankle angle. A very slight shift in the tracking position leads to a large vertical difference between the CMC position and the Inverse Kinematics position. The steady state error is very close to zero. I have included figures to demonstrate my point. The first is the CMC Perr output and the second is the comparison between the IK Kinematics and the CMC Kinematics.

My question is if it is acceptable for the Perr to go above 5degrees in this situation? if so, how much is acceptable?


I am also seeing large reserve actuators in the ankle, usually 15-25% of peek joint moment. The interesting thing is that they are always in the opposite direction of the peek torque, and are activated at the initial impact. Below is a figure that includes the ankle reserve, torque and GRF

I believe this may be due to the muscle dynamics, needing to provide such a drastic change in torque in a short amount of time. Does anybody have any clue to deal with this? I have tried the techniques presented in the documentation with minimal effect. Maybe I have not found the correct combination of parameters.


Both of these issues seem tied into one another, because when I am able to get one to budge, it changes the other. For instance the Perr could go to 15 degrees and reduce ankle reserve

I have also played with
1.Ankle tracking - interestingly, because there is not a lot of steady state error, increasing the weighting on the ankle has very minimal effect. I have had more luck decreasing the tracking so the ankle position is dictated more by the ankle reserve.
2. Ankle KP, KV - these parameters have large effects on the Perr term. In order to even get close to the recommended Perr values KP and KV need to be very large.
3. Decreasing the optimal reserve actuator torques- this helps minimally until I reach a threshold where the model launches off the ground.
4. Passive muscles forces, normalized fiber length and tracking weights.

Does anybody else have any experience with large forces and quick movements? It seems like this may be common issues.

Thank you for your time

[Ajay Seth]

There are no hard and fast rules about how good your kinematic tracking should be in terms of degrees of error with inverse kinematics results that fit the marker data in a least square sense. Ideally, you would compare your model marker results (with CMC kinematics) to the experimental marker data and the errors would be with the measurement error of the mocap system + skin movement artifact.

We provide some guidelines but they do not take into consideration high peak errors due to small phase shifts. If you compare your model kinematics to IK, it appears the CMC result is slightly lagging the IK motion and this is contributing to large peak errors. The dynamics of the model can lag due to filtering (joint coordinates and especially applied ground-reaction forces) and slow response of the controller (amongst other things). The CMC look-ahead (time) window of 10ms was selected for gait as a compromise between tracking accuracy and computation time. In your case, for a rapid movement I would consider using a window of 2 or even 1ms. It will make CMC a lot slower to run but it should generate a simulation where the muscle controls are more responsive and drive some of these peak errors down. Please let us know if this helps.

[Ryan Bakker]

Ajay, thank you for your response.

Decreasing the CMC look-ahead drastically improved my tracking errors for my high speed joints. My ankle pErr is now <.015 radians. Can you please explain what decreasing this window does? All i can find in the documentation is "A time window of 0.01 is generally sufficient for muscle activations to change enough to produce the desired accelerations." Is this because CMC is computed in chunks, or time segments? Decreasing the time segment would mean that it could change the control value more often, and therefor be more sensitive to changes? Are there any drawbacks to this method besides computation time?

I am still having issues with the reserve actuators. The reserve actuators are still behaving exactly the same as before. They are firing with significant force (30Nm) at the ankle, just not during peak joint torque and are firing in the opposite direction. This suggests the the model is "strong enough" but I may have another issue, maybe filtering or muscle dynamics. I am continuing to work through the steps suggested in the documentation. I have tried all of these steps with my previous solution but that solution had large pErr for some degrees of freedom. Any other suggestions for decreasing reserve actuators?

Thank you again for you reply,
-Ryan

[David Graham]

This is a fairly late reply as I was searching through the forum for another purpose when I came across your problem. Have you adjusted the segment masses of you model following RRA? The COM adjustment is made automatically but the segment mass adjustments are not and need to be done manually (as far as I know this is still the case). As such if you are working on a solution using the kinematic output from RRA but using a model without the suggested mass adjustments your residuals will reappear in the solution, thus might lead to the requirement to use reserve actuators. Might be worth a look.
Hope that helps

David