[ergoCubGazeboV1*]: model feet contacts are unstable

[martinaxgloria]

Trying to import ergoCubGazeboV1* models on gazebo, I spotted that the model was unstable, it was like wobbling on the feet. Together with @Nicogene, we used git bisect to understand which commit introduced the bug and it was the aae9cd6. We verified that the bug was not due to the introduction of the ft frames but the changes in the inertial values, actually some values have a different sign (see for example here)

Any idea on how to fix this problem?

cc @traversaro

[pattacini]

Similar to robotology/icub-models#209.

[traversaro]

Which version of Gazebo you are using?

[martinaxgloria]

Which version of Gazebo you are using?

I have the 11.10.2

[traversaro]

Related PR: #158 .

We verified that the bug was not due to the introduction of the ft frames but the changes in the inertial values, actually some values have a different sign (see for example here)

Good catch! Let me think a bit about it. For sure we could add a check on the physical consistency of the inertial parameters in the CI, by using the iDynTree::SpatialInertia::isPhysicallyConsistent() method

[traversaro]

We verified that the bug was not due to the introduction of the ft frames but the changes in the inertial values, actually some values have a different sign (see for example here)

Good catch! Let me think a bit about it. For sure we could add a check on the physical consistency of the inertial parameters in the CI, by using the iDynTree::SpatialInertia::isPhysicallyConsistent() method

Actually that is not the problem: the inertial parameter that changed sign is the com Z, and that make sense as it is expressed in the link frame, and we moved the link frame up, so the COM w.r.t. the link frame is now negative. However, it is not clear how this influences gazebo.

[traversaro]

I do not think the problem is in the COM location, by looking into the model, it seems that the collision of the links r_foot_rear, r_foot_front, l_foot_rear, l_foot_front are wrongly placed, and so the link that is actually colliding with the ground is r_ankle_2 and l_ankle_2 :

[traversaro]

And for completeness: the problem is that r_ankle_2 and l_ankle_2 bodies are not simple geometries, and we do not set their contact parameters in the .yaml file.

[traversaro]

I proposed a solution in #175, the PR also contains a (hopefully) detailed description of what happened, please comment if it is not clear.

Unfortunately #161 removed the simmechanics generation machinery, so to propose the change I created a ergocub-0.3 branch out of v0.3.4 and I opened the PR there. @Nicogene feel free to decide what is the best option to propagate this modifications in the master branch.

[martinaxgloria]

Thank you @traversaro!

[Nicogene]

I proposed a solution in #175, the PR also contains a (hopefully) detailed description of what happened, please comment if it is not clear.

Unfortunately #161 removed the simmechanics generation machinery, so to propose the change I created a ergocub-0.3 branch out of v0.3.4 and I opened the PR there. @Nicogene feel free to decide what is the best option to propagate this modifications in the master branch.

I can use the yaml you proposed with creo2urdf and publish the fixed urds