Kill JointTyped

[gchenfc]

This abandons future plans for Ball/Spherical/U-joints. Reasons:

• Gazebo doesn’t really handle ball joints fully anyway, since the different backends handle them differently and some don’t support them at all, so Gazebo just doesn’t fully support them, e.g. you cannot actuate ball joints.
• Previously I thought these would be useful to simulate cable robot, but I realized there’s a much better way.
• These types of joints (e.g. wrist, hip) are implemented with multiple “revolute” joints anyway, and due to joint limits don’t typically run into strange “singularity issues” (e.g. all joint angles = pi) anyway.
• Lots of other people just implement them as multiple intersecting joints, e.g. Lee09mmt which Frank shared (I didn’t review this paper in detail)
• The ball joints etc make lots of peoples’ lives (e.g. yetong) a lot harder.
• Therefore, I proposed we kill the complicated joint stuff and just revert back to only screw-type joints (joint angle has type double)

Frank’s response (after much deliberation):

• Ok
• Kill #286
• Gerry handle this new PR:
  • Kill big joint inheritance heirarchy
  • Basically, ScrewJointBase is the new Joint
  • Remove all joint templating madness
  • R-P-H joints only (H is ScrewJoint)
  • Minimal possible change to factors to make review easy

Can re-target to master if #290 is rejected. I forget why I based this branch on the other one - probably just forgot to checkout back to master before starting this work.

[gchenfc]

@dellaert I remember you said to wait on this until another PR landed (#278 ?) but not sure what the status on that is

[dellaert]

I think we can merge, @danbarla will have to merge in master then...

[gchenfc]

@dellaert Sorry I totally forgot to re-target to master, and then I royally messed up and accidentally force-pushed. But after messing up once, I learned so next time I'll know how to do it right.

But everything should be addressed now!