Hello,

I'm experimenting on collision free arm navigation with the KUKA youBot, and I have some strange problems that I would like to discuss with you.

I was able to play with the planning architecture in the planning scene warehouse: everything works excellently there, I'm able to plan and execute trajectories on the robot for a variety of poses (the only issue is a certain slowness in the trajectory execution, that I reported in another question).

However, strange problems occurs when I try to exploit the planning architecture from my code, using movearm. Using some code from the movearm tutorials, I get NOIKSOLUTION for almost all the poses that I specify, also for poses that where correctly solved in the planning scene warehouse. Sometimes I'm able to solve a certain pose, then I manually move the arm to another position and ask to get back to the previously solved pose, and NO_IK occurs! This non-deterministic behavious could be legitimate if using an iterative ik solver (dependant on the seed state), but the youbot-manipulation uses an analytical one... When the IK is able to solve, everything is ok.

My problem is similar, in general, to the one described here: http://answers.ros.org/question/10924/ik-works-in-planning-visualizer-but-often-fails-on-real-robot/ However, the problem, there, seems to be related to the iterative ik solver, while the youbot-manipulation relies on an analytical one.

Do you have any suggestion on that? I'm stuck on this problem!

More details: I'm using the youbot-manipulation stack from Kuka, on ROS Fuerte. The manipulation stack includes packages that are compliant to the planning architecture (youbotarmkinematics, youbotarmnavigation, etc)

Thank you