1
$\begingroup$

I have a question that lays on practical experience. What is the best adaptive controller for a robotic arm?

  • Self Tuning Regulator (STR)
  • Model Reference Adaptive Controller (MARC)
  • Adaptive Model Predictive Control (AMPC)
  • Iterative Learning Control (ILC)
  • Gain Scheduling (GS)
  • Auto-tuned PID (ATPID)

Notice that robotic arms are mechanical systems so the controller need to be fit into a microcontroller, e.g real time system due to the speed and robustness.

$\endgroup$
  • $\begingroup$ A control strategy is chosen always with a goal in mind for a specific problem. There is no general best approach, there is only the best suited approach for a given problem. Also..fitting into a microcontroller is a significantly larger challange then fitting into a real-time system. In practice (almost) all industrial robot arms use a cascaded PID loops. $\endgroup$ – 50k4 Jun 26 at 10:54
3
$\begingroup$

The given strategies can be divided into two groups: model-based and model-free control. The AMPC and MARC suggestions are model-based the other are model-free. Theoretically, a pid controller can be combined with Model-based control but for reason of simplification the assumption is, that each technique is located in a certain group.

Now we can ask if model-based or model-free is the more practical robotic arm controller. The good news is, that the winner can be easily imagined, because only model-based control is able to predict future state-space and this is equal to an increased accuracy. It is interesting to know, that model-based control is seldom teached in universities, because it can't be described mathematically elegant.

It remains one point open which has to do with adaptive control. Adaptive means, that the system learns it's parameters on the fly. It will allow to use the same controller for different robot arms. A heavy one and also a lightweight kinematic chain. The holy grail in adaptive control is a system which adapts itself to any kind of robotic arm, no matter if it's underactuated or fixed, with a low amount of DOF or with more servo motors in it. Creating an adaptive forward model is a bit complicated, and I'm not aware if this problem was already solved in the literature.

$\endgroup$
  • $\begingroup$ So you would say that an Adaptive MPC is the best choice for robotics due to the knowledge we can get from the model e.g prediction etc.? MARC is not 100% modelbased, sure, we using a model to only decribe the reference tracking. But MARC have its roots from the 60's. Soon 60-years old. Not sure if that technology is usefull today in modern applications. Who knows... $\endgroup$ – Daniel Mårtensson Mar 26 at 17:59
  • $\begingroup$ MPC is extensively taught at universities, I have yet to see an Advanced Control Systems or Advanced Control Theory course which does not teach MPC. $\endgroup$ – 50k4 Jun 26 at 10:50

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.