I'm developing a Task Space controller based on speed for a Universal Robot 5. I've already calculated the Jacobian, position and rotation error for the classic control method:

enter link description here

The method works fine in simulations (ODE Engine, Ros URDF), however, on a real UR5, I can't simply use the K parameters from simulations. From my experience, because the last three motors of the UR5 differ from the first three, the last three joint motors are much more sensible in relation to the value of the parameters in the K matrix. I managed to get to a 5 mm error in position with low parameters for the last 3 joints, but it's still unstable for long experiments (3+ minutes).

My question is, is there a method in the literature for optimal tuning of Task Space Controllers? How is it normally done? I couldn't find specific papers about it.

Thanks in advance.

  • $\begingroup$ What method did you employ to reach your initial K matrix parameters for your simulation? $\endgroup$ – koverman47 Apr 26 at 21:44
  • $\begingroup$ Is your $J^{-1}$ full rank? $\endgroup$ – koverman47 Apr 26 at 21:48
  • $\begingroup$ The Jacobian is full rank (I have also tested with pseudoinverse and DLS). The K gains in the simulation were found in an ad hoc manner. High enough to respond fast and have a small enough error and also be stable. That's because I couldn't find a way to do it optimally. $\endgroup$ – Felipe Henrique Apr 26 at 21:56
  • $\begingroup$ Would you be able to update your simulation to accommodate the distinct motor? If so, it would be much easier to test and there will be more tuning options available. $\endgroup$ – koverman47 Apr 26 at 21:59
  • $\begingroup$ I'm simulating in V-Rep. Last I checked this was not possible in Joint Mode, but I'll take another look. $\endgroup$ – Felipe Henrique Apr 26 at 22:03

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.