How does OMPL use sampling-based planners

edit

Hello ROS community,

I have a 6 DOFs manipulator robot. I read a lot of articles talking about motion planning with manipulator robots. I more or less understand the theory behind it and I've come to the conclusion that sampling based planners are best in my use case. I've come to this library called OMPL which implements much of the concepts in research. But something still intrigues me. The sampling in joint space and cartesian space is still a mystery :

1- How does OMPL sample in the cartesian space with the IK solver, considering for a specific (x, y, z) there's an infinite number of solutions in the joint space ?