There are many control methods who sounds great and have nice math etc. But many of them are not realistic in real life.

In robotics, Adaptive Control is well applied because Adaptive Control make sure that the controller is allways new and perfect for the robot.

But then there are another control method called Robust Control. In fact, Robust Control is only a design method. It's still a LQR/LQG/PID inside, but the parameters are designed so the controller has the best strength aginst uncertainties.

Then my question is: Is Robust Control well applied in robotics, or is Robust Control just another design method who does not work in reality, only in ideal simulation?


1 Answer 1


Robust control is used for when you know that parameters are going to fall within particular bounds. For example, for a mass-produced robot arm, the arm masses fall between X and Y, the moments of inertia between I and J, etc. You don't know exactly what they are, but you can analyze the boundary conditions (i.e., root locus analysis) and ensure stability and performance parameters are met at all times.

Adaptive control is used for when you don't know that the parameters are going to fall within particular bounds. Maybe you've got a kite controller, but the wind can be anything between zero and a hurricane, or an autopilot system in an aircraft where the total passenger and equipment weight isn't known.

A robust controller is developed and analyzed once, and then the gains are fixed. An adaptive controller must constantly assess the system performance and attempt to adjust control gains on-the-fly. The adaptive controller must be a more computationally complex controller (for comparable systems) because it cannot leverage a priori knowledge like the robust controller.

Ultimately, the best controller is the one that performs the best for your system. I would argue that robust control would probably best be applied to robots working in controlled conditions, or fabricated of mass produced parts, and adaptive control then applied to robots working in uncertain environments.

  • $\begingroup$ One problem here for me is that I can't be good at both robust control and Adaptive Control at the same time. Those are huge techniques of control engineering. Learning a little bit of them both is not an argument to say I know the theory behind. But I need to select one. $\endgroup$
    – euraad
    Jan 16, 2018 at 21:39
  • $\begingroup$ @DanielMårtensson - I would argue that the first step in deciding to utilize adaptive control would be demonstrating that traditional control (pole placement, etc.) is insufficient. A robust controller has fixed gains, and an adaptive controller has to utilize some feedback mechanism to self-tune. You need to be comfortable with designing controllers in general before you move on to adaptive control. To me, this is along the lines of saying something like, "I can't be good at both algebra and calculus at the same time." One is kind of the foundation for the other. $\endgroup$
    – Chuck
    Jan 17, 2018 at 14:54
  • $\begingroup$ So you mean that I can be good at both Robust Control and Adaptive Control just by knowing the controllers structures? For example, in Adaptive control there are three main controllers, MRAC(Model Reference Adaptive Control), STR(Self Tuning Regulator), MVSTR(Minimum Variance Self Tuning Regulator). But knowing the application of those are much larger that the controllers it self. When I read my book about Adaptive Control(Åström), I don't know what is necissary to know and what is not necissary to know. That makes control theory difficult. $\endgroup$
    – euraad
    Jan 17, 2018 at 22:01
  • $\begingroup$ @DanielMårtensson - Yes, you can be good at both! I would suggest you read more about model based control, like state feedback controllers, and read more about methods of system identification. I think you'll recognize what I'm trying to convey here - A "robust" controller is a conservative model based controller, and an adaptive controller is a model based controller that also performs some version of system identification. The different "flavors" of adaptive control are generally different by how and when they perform system identification. $\endgroup$
    – Chuck
    Jan 17, 2018 at 22:13
  • $\begingroup$ For instance, you could view observer design, Kalman filters, etc. as a form of adaptive control. Classic observer design uses a model as a basis for predicting or filtering feedback (state variables), but you can use the same techniques for estimating parameters (entries in the state matrix A). Once you're estimating parameters, presto! You're doing adaptive control. $\endgroup$
    – Chuck
    Jan 17, 2018 at 22:15

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

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