Path plannig using MoveIt in cartesian space with non-unique inverse kinematics

edit

I am using ROS1 Noetic and MoveIt for path planning of a custom-defined robot.

My main challenge that I have is that the end-effector target pose is not completely defined (the rotation around the z-axis is arbitrary). The robot is moving along a rail, and a rotating, telescopic arm is used to reach a certain destination. I do not really care about where on the rail the robot is as long as it can reach the target; if the target is too close, I have to move away along the rail until the minimum distance is reached. But it is not clear where I should move on the rail since there can be obstacles in the way. And depending on where I am along the rail, the rotation of my end effector is different along the z-axis when reaching the same position.

I'm sorry I am not yet able to upload images to explain things a little better...

I have written an inverse kinematics which calculates the joint values for a specific target position and a given position along the rail; for such a combination, the solution is unique. I have also tried writing a InverseKinematics plugin for the MoveIt library and succeeded building it.

But I am neither sure how to correctly implement it since the solution is non-unique so long as the position along the rail is not fixed, and I am not sure how to correctly define the target of the path planning task:

• If I use joint-based path planning, I have to decide up front where on the rail I should end up, and then providing a target pose in joint space is easy. However, I need more intelligence to determine the desired position along the rail.
• If I provide an end-effector target pose, I need to also specify the orientation of the end-effector, and by specifying the rotation around the z-axis, the position along the rail is again uniquely determined (which I do not want).

What I would like to do is let the path planner decide on the ideal final position along the rail (probably even using a metric that favours positions close to the initial position). Along the same lines, I would like to ask the MoveIt library whether a certain point can be reached at all, from any position along the rail and without collisions.

I hope I could make my problem sufficiently clear and look forward to your ideas! Thank you!