The "standard" model against which ligament scaling is tested contains correct path points but incorrect `resting_length`s

[tkuchida]

I provide a detailed description of the issue below because the resolution involves changing a "standard" test model.

This issue concerns the following files in the Applications/Scale/test directory:

• testScale.cpp
• toyLigamentModel.osim (the original model)
• std_toyLigamentModelScaled.osim (the "standard" model)
• toyLigamentModel_Setup_Scale.xml (the ScaleTool setup file)

In testScale.cpp, line 244, the scaleModelWithLigament() method uses the ScaleTool (with the toyLigamentModel_Setup_Scale.xml setup file) to scale the "block" and "block1" bodies in toyLigamentModel.osim by (2, 2, 2). The "standard" model (std_toyLigamentModelScaled.osim) contains the correct GeometryPath points, but the resting_lengths are incorrect. Here are the values of the resting_lengths in the original and "standard" models:

Table 1

Ligament	resting_length in original model	resting_length in "standard" model
"lateralLigament"	0.14	0.14
"MedialLigament"	0.14	0.14
"internalLigament"	0.04	0.06
"crossLigament"	0.147	0.289196876423888

The Ligament's postScale() method indicates that the resting_length should be multiplied by scaleFactor = {path length after scaling} / {path length before scaling}, so the table above suggests that "lateralLigament" has not changed length; however, this is not the case.

I opened the original model and "standard" models, and used the following script to read the lengths of the ligaments in the default pose:

model = getCurrentModel();
s = model.initSystem();
modeling.Ligament.safeDownCast( model.getForceSet().get(0) ).getLength(s);

Here are the lengths:

Table 2

Ligament	Length in original model	Length in "standard" model
"lateralLigament"	0.14	0.28
"MedialLigament"	0.14	0.28
"internalLigament"	0.04000000000000002	0.08000000000000004
"crossLigament"	0.1469693845669907	0.2939387691339814

I drew a sketch of the bodies and frames, and confirmed that "lateralLigament" should have doubled in length as a result of scaling the bodies by (2, 2, 2). Ligament's postScale() method should have updated the resting_lengths by computing scaleFactor = {path length after scaling} / {path length before scaling}; in the case of "lateralLigament", scaleFactor = 0.28 / 0.14 = 2 (per Table 2) and resting_length should have doubled (not remained the same length, as shown in Table 1).

Not sure why the ASSERT on line 291 doesn't currently throw, but I'm confident that the "standard" model is incorrect.

[tkuchida]

Here's a screencap of the ligaments in the original model (blue) and "standard" scaled model (green). None of the ligaments remain the same length.

[chrisdembia]

I don't see an issue with updating this standard. Is there anything from the archaeology to suggest otherwise?

[tkuchida]

Is there anything from the archaeology to suggest otherwise?

I found a comment from @aymanhab in #538: "The test case is validated against manual inspection for a simple block with attached ligaments. There's no existing code that would have scaled ligament resting length properly, the scaling of geometryPath for this model is trivial and has been visually verified by @jimmyDunne and myself." I'm investigating on master now…

[tkuchida]

On master, the correct scale factors are calculated (and, consequently, testScale fails) if s = _model->initSystem(); is inserted in SimbodyEngine.cpp after scaling joints. Without initSystem() here (i.e., currently on master), Ligament gets incorrect path lengths in its postScale() method. The lengths are correct after Model::scale() returns (presumably because of the s = initSystem(); on line 1504), as reported using the 3-line script shown above. I haven't been able to determine exactly what's going on (possibly outdated cache, e.g., in GeometryPath::computePath()?), but I think the current "standard" file is incorrect and hypothesize that it was generated using OpenSim rather than analytically.