simtk.org

I have seen the MOCO user guide, discussions in this forum and papers (Miller 2021 PeerJ) that suggest replacing other muscle models with DeGrooteFregly2016Muscle in order to work with MOCO. The reason for this is that DeGrooteFregly2016Muscle has all characteristics (FL, FV curves) C2 continuous (tendon FL C3 continuous), which meets the requirement of the gradient solver used by MOCO. Also, the implicit formulations of contraction dynamics implemented in this model may converge the better than explicit formulation.

The Millard2012EquilibriumMuscle model will sort of work in MOCO similar to the second formulation (explicit formulation with fiber length Lm as the state) in DeGroote 2016 paper, as all characteristics are C2 continuous. However, this formulation will be hard to converge as demonstrated in DeGroote 2016.

The Thelen2003Muscle model will likely not work, because it does not have characteristics that are C2 continuous according to the OpenSim Characteristic Musculotendon Curves page. https://simtk-confluence.stanford.edu:8 ... don+Curves

Is this understanding right? I am just starting to get into MOCO, and want to get this right. Thanks!

Hi Xiao,

That all looks right to me! Some of the "non-smooth" models might work in certain situations, but the DeGrooteFregly2016Muscle model should be the most reliable/well-behaved model, since Moco utilizes gradient-based optimization which requires smooth, continuous functions.

Best,
-Nick