May someone, in simple terms, describe to me the difference between a Kalman filter and a linear quadratic regulator? If possible, please use an analogy or maybe even a visual demonstration of the difference. Also, some basic mathematical descriptions would also be helpful...thank you.

  • \$\begingroup\$ I'm not a DSP expert so I can't fully answer. Kalman filters are very broad, and just use the concept of "state space". Many, many linear adaptive filters can be derived from the Kalman (such as the RLS algorithm). I'm not very familiar with the linear-quadratic regulator, so I can't help you there. \$\endgroup\$ – Joren Vaes Apr 25 '17 at 5:36
  • 2
    \$\begingroup\$ The Kalman is a filter, while the other is a controller. \$\endgroup\$ – Marko Buršič Apr 25 '17 at 8:32

The LQR is a linear quadratic regulator (controller) and can be applied if the system is fully controllable. It takes the full state vector and transforms it into an input signal to achieve desired properties of the dynamical system. In many practical problems we are not able to measure the full state vector (e.g. sensors are too expensive). In order to estimate the full state vector we must check if the system is observable (e.g. by Kalman's observability criterion), if the system is observable then it is possible to estimate the full state vector by a Kalman filter (LQE = linear quadratic estimator) and use this estimate of the full state vector as input for the LQR controller. Note that the combination of LQR and LQE, sometimes also called LQG (Linear Quadratic Gaussian) might be very sensitive to model uncertainties. This means that even very small uncertainties can lead to an unstable LQG.


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.