We are running a project that includes force-based control with a KUKA iiwa. Therefore we need joint position, velocity and effort readings as well as the feedback from the FT sensor plugin.

Using Gazebo from the ROS repositories for melodic gives us wrong readings for joint velocities, joint efforts and the FT sensor plugin when being in singularities or close to them. An example for the values can be found here:

This is an iiwa going from some start position into the home position with all joints being 0. For this robot, this is a singularity for joint 1, 3, 5 and 7. A similar behavior can be shown when only bringing two joints into a singularity. In this case, the trajectories are generated with MoveIt and the execution looks okay. We are using the ROS control with the PositionJointInterface, but we specify the gazeboroscontrol PID gains.

These readings happen both when using jointstatecontroller and the Gazebo ROS joint state publisher plugin. Whereas the latter one does not publish efforts since it is not included in the code of the plugin.

The same time the joint velocity and joint effort readings are off, the FT sensor values are off as well.

Besides switching between jointstatecontroller and the Gazebo ROS joint state publisher plugin, I also compiled Gazebo from source. These are the results:

Gazebo tag gazebo9_9.16.0 and ODE for physics:

Same behavior as Gazebo from the package repository.

Gazebo tag gazebo9_9.16.0 with dart tag release-6.9:

Same behavior as Gazebo from the package repository.

Gazebo tag gazebo9_9.16.0 with bullet 3.07:

The robot spawn partially inside the floor. I haven't looked into why this happens. But I am also not really interested in Bullet right now.

This behavior for can be reproduced with iiwa_ros and its Gazebo launch files. I'd be grateful for some pointers into the right direction. Both ODE and Dart would be options if some solution works in any of them.