simtk.org

I can tell OpenSim to use CSFQP as its optimizer during simulations when running things from the command line and by manually editing xml settings files, but is there a way to tell OpenSim to use CSFQP by default instead of using ipopt when using the GUI? Does it automatically use CFSQP if it detects the library in the \bin folder?

Thanks!

Hi Chris,

If you load that same settings file you edited by hand in the GUI, even though it's not exposed, it should run that analysis using CFSQP. You can check by watching the messages window. When the optimizer is loaded it will tell you if it is using CFSQP or IPOPT.

Hope that helps,
Sam

Hi Chris & Sam,

We don't want to make it the default if you have the library since having the library does not mean that it is your preferred choice.

Long term, we should be able to query for available Optimizers and show you a dropdown in the GUI. The support for this in the XML file is already there as Sam pointed out.

Just curious, which problem did cfsqp perform better than ipopt on?

-Ayman

Right, but is there a way to make it so I don't have to edit each settings file, or choose cfsqp even if I'm not using a settings file at all and am just running things in the GUI on the fly?

Sorry about that last message Ayman, I sent it before you saw your response.

I'm getting roughly 25% faster IK simulations with CFSQP over IPOPT (via command line) and virtually identical results. My simulations are taking hours, so 25% is a lot...

Chris,

We're looking into IK performance in particular. One thing I found out today is that if you take out the muscles completely (they are not used for IK anyways unless you're displaying them in the GUI), the time is cut by half for the gait2392 example. We'll fix this soon on our side, but just something to consider. When animating results later you can use your model with muscles.

Just a quick thought..
-Ayman

How would one do this? Do you mean have two identical models, one with the muscles and one with them removed, and just apply the kinematics calculated via the muscle-less model to full model? It does seem odd that removing muscles would alter performance since they (theoretically) aren't doing anything for IK simulations...

Chris,

Yes, it's an acknowledged bug, especially since you're running from the command line and that we'll fix soon. And yes you can run IK on the muscle-less model and show them on the model with muscles.

-Ayman

Is there an update to the CFSQP.dll? The osimCFSQP.dll file I started using win Osim2.1, or so, does not work when called by 2.4 or 3.0 beta.