IK solutions for the robot don't use a seed state at all

[gaya-]

...which was a result of the switch from arm_kinematics tools to moveit, namely cbb8ac0#diff-bd86881dfc36ce9675b4cfb5fa238808R289

[airballking]

We still have the old arm-kinematics code around if you need to get the solution with seed-states back.

[gaya-]

Oh no, no way, going back to arm_kinematics would be a stack backwards. We'll take a look with Mihai into this and try to get moveit accept seed states.

[fairlight1337]

Hey,

I already looked into this - just to get it right, this gives the robot a hint where to look for the solution, correct? cram_moveit right now includes a hacked-in capability of raising the elbow when doing manipulation (to not collide with the table), but I guess it lacks generality.

[gaya-]

Hey, I just looked at cram_moveit, namely here, but couldn't find anything with the seed state. Did I get it right that you're using joint constraints instead? That's a good idea as well and should definitely help with the elbow. But I'm not completely sure if the elbow is the only reason why finding a good IK solution is so hard for moveit right now, at least in my setup. Still haven't found time to look into this...

[fairlight1337]

Hey,

right, I'm using the joint constraints. Haven't found a way to set a seed state either, that's why this solution ended up in the code (although still dirty as it is). If I see something that does the job better (i.e., allowing other solutions that would violate the constraints and only take an initial hint about where to look), I'll introduce it.

Suggestions welcome :).