Fix HumanStateProvider on Windows

[lrapetti]

We have tried to run the HumanStateProvider and the visualizer on a Windows machine, and we have noticed a strange behaviour.

Results are shown in the following video where we run the HSP+visualization on the same dataset and same configuration (Human.xml):

video6051136142948960770.mp4

There are two main problems:

• (A) IK output seems to be wrong
• (B) When we stop the data, we expect the IK output to be fixed, instead it keeps changing.

Some possibl source of problem could be:

• Visualizer (it seems not since it wouldn't explain problem (B)). But we can easily double-check by running HSP on Windows and visualizer on another machine
• Some non initialized matrices. In this case running in debug the problem shouldn't happen
• IK parameters tuning

[S-Dafarra]

• Visualizer (it seems not since it wouldn't explain problem (B)). But we can easily double-check by running HSP on Windows and visualizer on another machine

The reverse (Visualizer on Windows and HSP on another machine) could be interesting too.

[traversaro]

Is this using vcpkg or conda dependencies? How is the locale set in the different machines?

[lrapetti]

Is this using vcpkg or conda dependencies? How is the locale set in the different machines?

On top, I am on MacOS, dependencies installed via homebrew and human-state-provider installed via robotology-superbuild. Same behaviour is experienced on other machines on Ubuntu (apt+superbuild).

On the Windows machine (visualization on the bottom), dependencies are installed via vcpkg and human-state-provider via robotology-superbuild. (please @kouroshD correct me if I am wrong)

[traversaro]

Same behaviour is experienced on other machines on Ubuntu (apt+superbuild).

The same behaviour described in this issue?

[lrapetti]

Same behaviour is experienced on other machines on Ubuntu (apt+superbuild).

The same behaviour described in this issue?

No, on Ubuntu we have the same behaviour that we experience on MacOS.
The problem appears to be only on Windows.

[traversaro]

If the functionality is part of a released version of human-dynamics-estimation, a quick check you can do without recompiling the world is to just install human-dynamics-estimation via conda binaries on Windows:

mamba create -n hde-env
mamba activate hde-env
mamba install -c conda-forge -c robotology human-dynamics-estimation
# Make sure that all environment variables are properly set
mamba deactivate hde-env
mamba activate hde-env

If you are already in machine where the robotology-superbuild with vcpkg is installed, make sure that you clean the enviornment from any globally set env variables, using the appropriate powershell scripts.

[kouroshD]

How is the locale set in the different machines?

@traversaro what do you mean by locale here? Is it related to https://docs.microsoft.com/en-us/windows/win32/intl/locales-and-languages ?

[traversaro]

How is the locale set in the different machines?

@traversaro what do you mean by locale here? Is it related to https://docs.microsoft.com/en-us/windows/win32/intl/locales-and-languages ?

Yes, kind of. I asked that as there are a lot of bugs related to the use of functions that convert string to floating point numbers (and back and forth). See robotology/idyntree#288 . See https://serverfault.com/questions/173630/how-to-get-system-locale-in-windows-7-cmd for how to get it.

[lrapetti]

• But we can easily double-check by running HSP on Windows and visualizer on another machine

We did this test and visualization still seems to be wrong, so we can exclude it is a problem of the visualizer.
We also noticed that the human-state-provider on Windows, from time to time was starting printing the warning for velocity out of the limit, with huge powers.

[lrapetti]

This problem has not been reported for a while, and I know that the HumanStateProvider was running on windows on @LorenzoFiori and @AntonellaTatarelli laptop.

I will close this issue, we will re-open it in case we experience the same problem again.