Here are my experiences using joint modes:
Disengaging
Disengaging after a certain amount of time works the best. If your joint is actuated then disengaging/being disengaged does not work, like you describe. I use a quite good workaround where I place a "Weld Joint" before the actuated Joint and disengage that. In practice it is the same result.
Mode: Provided by Input | Step from 0 to -1Engaging
When starting out as disengaged, the placement of the follower part is not really determent, since the joint is disengaged. When testing the "floating"-body was placed with the R-coordinate system at the world-origin. The moment you are trying to engage the base and the follower frame should have the same orientation and position (and velocity?). Otherwise the following error message gets displayed:
Error:An error occurred while running the simulation and the simulation was terminated
Caused by:
['modelName/Solver Configuration']: The position components of the kinematic constraints ...
of the mechanism cannot be satisfied when attempting to transition joints from disengaged ...
to normal mode. This is more likely to occur for large rotation errors compared to large ...
translation errors. Consider reducing the error in those joints before attempting the ...
transition or consider increasing the joint mode transition nonlinear iterations parameter ...
in the associated Mechanism Configuration block.
By testing with an ideal example along one axis, I found out that an angle error of about 10° is working fine and I found no upper limit on translational displacement. This changes when an error occurs on multiple axis, but I did not investigate further. Less error should result in a higher chance that it works though.
I attached an example for engaging and then starting to rotate, which is very simple. But to be honest I did not get it to work with a more complicated setup. Still, I recommend trying.
