I am modeling an elastic ball joint (composed of two revolute joints, because I am using urdf) and have an issue, where Joint angle values reported from gazebo seem to fluctuate from negative to positive for small angles at seemingly random. I saw a similar issue mentioned here before, but without a good solution or means to debug: https://answers.gazebosim.org/question/21286/physicsjointgetangle-returning-incorrect-values-close-to-0/

The picture looks relatively similar (taken from plotjuggler, y axis is joint angle in radians, x axis is seconds). Note that I am using a jointstatecontroller from gazeborospkgs to publish these values, looking at it's source it calls the newer Joint::Position method, but issue looks similar. This picture is taken after moving the platform in a direction parallel to x axis (first 17 seconds), then moving it parallel to y axis (angle at 0.1 radians) and then moving it in a few other directions.

I am using these elastic joints for sensing the environment, and also need them to be accurate on small angles. What can I do to get joint accurate angles? Should I try to calculate them also from world pose of the joints, like done on the previous question?

I am able to share the project, if someone more capable is interested to take a look at why this is happening. I've already added the model to the post (it is actually urdf, but renamed to sdf, because I cannot upload an urdf). model

Also added the whole project to git, in case that is easier to look at: https://github.com/AndreasNagel/get_angle_issue_sample