A few beginner questions about Moco

[Aaron Cherian]

Hi there,

I'm new to using Moco and I've been through the documentation and some of the example files, and after going through those I had a few questions I wanted to ask. For context, I'm primarily working with 3D gait data, so I was looking through the MocoTrack and MocoInverse examples for 3DWalking in Matlab.

First, what is the main advantage of MocoTrack? Does the addition of the cost error make the results from MocoTrack more reliable or accurate than what we get from MocoInverse?

Second, when thinking about the workflow, I noticed that for muscle-driven tracking and inverse example problems, the example files reference a 'coordinates.sto' file, but it's not clear where this coordinate file comes from. I noticed that after running the RRA tool on the data that I was working with that there was a file named '3DGaitModel2354-RRA_2_Kinematics_q.sto', which had the same 'Coordinates' heading on it as the 'coordinates.sto' file used in the MocoTrack/Inverse examples. Is that '3DGaitModel2354-RRA_2_Kinematics_q.sto' file the one that's being referenced in the MocoTrack/MocoInverse examples? If so, does that mean that you need to run the RRA tool on your data before using Moco, or is there another way to generate that 'coordinates.sto' file?

Third, what would the best way be to check that your Moco results are 'good', so to speak? Are there error values of sort that we can reference?

Thanks again for the help.

[Carmichael Ong]

Looks like you're already getting deep into Moco! Some answers and pointers below:

1. MocoTrack and MocoInverse have different purposes, since MocoTrack "tracks" (but does not match) the input coordinates, but MocoInverse does ensure that they match. In general, this means MocoTrack will likely take longer to solve, so if your research question can use MocoInverse you should tend to use that (e.g., you want to know muscle activations given specific date), and if it's not flexible enough then MocoTrack may be helpful (e.g., you want to know how muscle activations might change if your kinematics can deviate from data slightly).

2. The coordinates will often come from performing an Inverse Kinematics step from marker data. They can also come from RRA (which takes IK data and then changes the model mass properties and kinematics slightly to be more dynamically consistent).

3. "Good enough" will also depend on your research question. The Validation and Verification paper could be a good place to first get some ideas on how to think through your project and the validation (https://nmbl.stanford.edu/publications/ ... ks2015.pdf).

By the way, in case you have more technical Moco-related questions, the OpenSim Moco Forum may a great place to post questions as well!