simtk.org

Hi all,

I have a hand model with measured small inertial properties (1E-7). I have a problem running CMC during finger flexion. it gives vey large position errors. Multiplying the properties by 1000 solves the problem .
I tried to solve the problem by having passive forces with coordinateLimitForces for the joints. but if I have it for all three joints MCP ,PIP and DIP it doesn't proceed to the next step. having one passive force at MCP make the process very slow.
I am wondering if it intrinsic to Opensim hat cannot handle samll inertial.

P.s: static optimisation works fine.

Thanks in advance,
Mojtaba

Two points:

1)
If the inertia or mass is very small this can cause problems when inverting the inertia mass matrix (e.g. think of it as a division by zero). So you have to be careful how to define the mass properties. This is not an inherit problem of OpenSim, but a numerical issue.

2)
About passive forces. If your passive forces are large then during forward dynamics (numerical integration), you are applying large forces to bodies with small mass properties. This causes the integrator to take smaller steps to predict the actual movement. You are making your problem "stiff". I am not sure how the passive forces helps you solve the first problem.

In contrast Static Optimization works because no numerical integration is required.

Thanks for the reply.

So you do not have any idea to solve the issue?

Yours,
Mojtaba

Find the inertia parameters (mass, inertia, center of mass) and define them correctly (look at alternative models or publications that report them). Don't let a body have zero mass properties.

The imported values are measured and make sense .
Another colleague of mine had the same problem with foot toes and he had to scale them to be at least1E-4 to make CMC work.
So I guess this is a mathematical issue.

Hi Mojtaba,

You could try Friedl her method for solving the muscle redundancy problem (https://simtk.org/projects/optcntrlmuscle/).
This direct collocation method does not invert the mass matrix.
But be carefull, we have only tried it for the lower limb muscles during walking and running.

regards,
Maarten

Hi Mojtaba
I am also performing CMC of hand model. I have used the mass and inertial properties from research paper benjamin et.al (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5597339/) but since mass and inertial properties are smaller (10^-7 range) so I am also getting too much positional inaccuracy same as your problem.

Have you been able to reduce the positional inaccuracy ? Any help in this regard is highly appreciated.

Thanks
Ankit