Model in different physics engines Gazebo 4.0

[hu-yue]

I open a new issue regarding the model of iCubHeidelberg01 that was obtained from a direct conversion from CAD as a reminder of the encountered problems. @traversaro and I tried to load the model in Gazebo 4.0 with different physics engines (ODE, Bullet and SImBody), the model was loaded correctly and if the one being loaded is the fixed one it seems there is no big difference among the different engines. However if we take the model that's supposed to stand on the ground, the following happens:

• ODE: seems ok with appropriate PID gains
• SimBody: it's as if the ground doesn't exist and the robot goes through vertically, this seems a bug of Gazebo or SimBody and not as the issue #126
• Bullet: with the same gains as ODE the robot cannot stand, it might be friction problem or PID gains problem? Is this related to issue #128?

[traversaro]

For SimBody : I guess that is caused by Gazebo Issue https://bitbucket.org/osrf/gazebo/issue/883/simbodycollision-doesnt-support-meshes-or . I suggest voting for this issue, however for getting simbody to work it could also make sense to substitute collision meshes with primitive shapes (useful also for performance reasons).

For Bullet: could it still be related to friction problems, I would consider looking if we define all the friction parameters that the atlas model is defining: https://bitbucket.org/osrf/drcsim/src/2df318b039f040dca6b430534dd104abd7a23cc7/atlas_description/urdf/atlas_v4.gazebo?at=default .

[hu-yue]

In the sdf format there is the possibility to specify contact and friction parameters for ode and bullet (http://sdformat.org/spec?ver=1.5&elem=collision#bullet_friction). In icub-gazebo there is only ode, however I tried to introduce the bullet parameters in iCubHeidelberg01 and it doesn't seem to work, but I need to check further if the numbers I put make sense and if it is ok to put both ode and bullet.

[traversaro]

I found that most of the instability problems are related to the "fake" fixed joints thay we are using for simulating ft sensors. I guess that until this bugs in Gazebo are fixed, it could make sense to provide a version of the model without this "fake" fixed joints for increased stability. This will be trivial using the URDF format and the <disableFixedJointLumping> joint, as soon they merge my PR https://bitbucket.org/osrf/sdformat/pull-request/126/added-urdf-gazebo-extension-option-to (it then should be available in Gazebo 5).