Large RMSE and Question about OpenSim ID

[Duncan Lee]

I am getting large RMSE from addBiomechanics (3.8 cm) and am wondering how to improve this. I am uploading 15 30-second level-ground walking trials all of the same subject, but each individual trial also has a large RMSE. The IK motion loaded into OpenSim appears to walk smoothly.

Also, when I try to run Inverse Dynamics in OpenSim I get ankle moments that do not look correct to me. The values are out of range that I would expect in Nm and the curves themselves are noisy. I am running ID using the skeleton from AddBiomechanics, the IK from AddBiomechanics, and force plate GRF data filtered with a low-pass butterworth.

ID_ankle_moment_err.jpg (125.46 KiB) Viewed 884 times

All data is captured using a Vicon system and Bertec force-enabled treadmill. Any help is greatly appreciated!

Thanks!

[Nicholas Bianco]

Hi Duncan,

The noisy ankle moments from ID suggest to me that you might need to filter your inverse kinematics joint angles, revise how you're filtering you GRFs, or both. The large RMSE could have several explanations, but usually it boils down to inconsistency between kinematics and kinetics in the model.

Without any more specific information, as a first step I would recommend taking a closer look at your raw data and try visualizing the filtered GRFs along with the filtered IK solution to check for consistency.

Best,
Nick

[Duncan Lee]

Hi Nick,

Thank you for the prompt response! I used a residual analysis to determine the cutoff frequency for my filter, and have included an example plot below.

Do you have any recommendations on filtering the IK data? As in, should I filter the raw TRC file or the output from AddBiomechanics? When I plot only the joint angles the results look smooth. I have tried using the OpenSim ID filter data option at 6Hz, but did not see any significant change when I plotted the moment.

Thank you!

[Nicholas Bianco]

Hi Duncan,

It would filter kinematics after running IK, and as a starting point, try using the same cutoff frequency as your ground reaction forces.

My guess is that filtering is not the primary issue here. Are you running the dynamics fitting step in AddBiomechanics? You could try running only the scaling + inverse kinematics steps and then run inverse dynamics in OpenSim to see what you get.

If the the inverse dynamics moments still look bad, then there might be something else wrong with the model/data. As I said before, visualizing the data in the GUI is a good first step to check for kinematic/kinetic consistency. In particular, check that the point of application of the ground reaction forces relative to the foot markers looks good. Small errors there can lead to large moment errors.

Best,
Nick