Revision history [back]

I am not 100% on this, but to me this seems like a problem i had with moveit constrained planning in the past as well. My guess: This is caused by "jumps" in the IK, where each sampled point is individually fine, but the interpolation between them leads to these "flips". This would describe why some segments of the trajectory look just fine, but the transitions between them are so messed up. Some discussion on this:

• This issue describes the problem
• This old mail thread discusses some ways to work around this, among others changing/parameterizeing the IK algorithm used.

One sure way to get rid of this should be using position-only ik as described here, but this can drastically increase your planning times. Changing the range parameter in config/ompl_planning.yaml was less drastic, but still helped in my case. One simple thing to help with the increased planning time was relaxing my orientation tolerances, e.g. in my case i decided that a tolerance of 0.3 was more then enough.

Could you try validating that this is the problem (i.e. checking if position-only ik / the other approaches listed here improve your results)? If this is the case it would also be interesting to hear what change could bring the biggest improvements for your setup.