MocoStudy fails after parameter optimization

[Christian Greve]

Dear OpenSim Moco community,

I am testing the effect of parameter optimization on muscle force computations during gait. The problem I encounter is that after- optimizing tendon lack and opt fiber length Moco does not find a solution. With the scaled-only model it does find a solution. The solver settings are exactly the same for both approaches.

When generating muscle lengths and activations of the failed solution I found that muscle activations are high at the first frames of the analysis in the solution with the optimized but not the scaled-only model. Also one specific muscle is having activations at 1. Since the normalized muscle fiber length and velocities are well in the recommended boundaries (0.6 - 1.4 norm fiber length) in the solution of the optimized model I do not understand why this is happening. To add, this is not a coincidence. The simulation with the optimized model fails across different attempts and experimental trials.

My questions are:
1. Do you have any suggestions what I could check on to solve this issue and get the solution with the optimized model succeeding?
2. Is it possible that the high activations at the start of the simulation and/or the muscle with activation being 1 are causing the simulation to fail?
3. Is it possible to somehow control for the high activations in the solution with the optimized model file?

Attached are some plots for clarification.

Thanks and kind reards

Christian

[Nicholas Bianco]

Hi Christian,

Are you using the DeGrooteFregly2016Muscle? If so, this is probably due to a known bug where some of the don't get updated properly when performing a parameter optimization problem. I've been meaning to fix this for a while, so thank you for the reminder!

Creating an issue now on GitHub to keep track of this.

Best,
Nick

[Christian Greve]

Dear Nicholas,

thanks for your fast response. Yes I am using "DeGrooteFregly2016Muscle" but I am not sure that 'not updating' is the problem.

I first use parameter optimization to define new values for tendon slack and opt fibre length--> then generate a moco solution with MocoInverse. If I use the optimized model, MocoInverse fails. But If I use the scaled-only model it finds a solution.

So I would think it is not in the parameter optimization. Could that be?

Regards

Christian