use moveit! computeCartesianPath to plan a circular path

edit

thanks for your reply. I already use a robust ik solver for that. I have a thought that if I take enough waypoints on the arc path, and solve the ik for each waypoints, thus get a trajectory of joint values. Then use iterative_time_parameterization to add the timestamps. Not sure if this works, will try tomorrow.

FWIW, what you described above is basically exactly what computeCartesianPath does (see source here). With one minor addition -- there are several heuristics to determine if you've accidently found IK solutions at neighboring waypoints that have large "joint jumps" that might not be feasible in real life. You could experiment with the values used for joint jump testing. Also note, it's quite possible that even with "robust IK" there doesn't exist a solution that follows your desired arc while also changing orientation (however you've defined that), or perhaps the only solutions that do exist require large joint jumps. An obvious cause that could produce this scenario is joint limits.

thanks very much. Indeed, I tested the way mentioned, the main problem is the jumps. But I'm still a bit curious: since the last joint is free to rotate along the axis (updated in the figure), it should have an ik solution without jump on the waypoints where computeCartesianPath succeeded when I keep the orientation same with the initial one.

solved... the problem is that the eef joint needs to move from pi to -pi, leading to a large jump. sorry for my mistake..

I converted your comment to an answer and marked it as correct. That way this question shows up as properly answered. Glad you were able to figure things out!