Hi Petru-Daniel,

the answer to this question depends on the type of MPC controller that you want to implement. I guess you refer to MPC for linear systems and using quadratic cost functions? The computational effort to solve the MPC problem for linear models and quadratic costs depends on the constraints that you are considering. If you consider a linear model, a quadratic cost function and affine input and state constraints you end up with a quadratic program (QP) to be solved online. Roughly speaking the QP gets more complicated the more complicated your constraints are. And input constraints are usually easier to handle than state constraints.

If you only consider simple input constraints you might find the following publication interesting:

https://www.researchgate.net/publication/259671909_Implementation_Aspects_of_Model_Predictive_Control_for_Embedded_Systems?ev=prf_pub

In this paper we have shown that you can implement an MPC controller even on a LEGO mindstorms platform. And this platform relies on an ARM processor. If you want to consider state constraints you need more sophisticated algorithms since the underlying quadratic program will be more complex.

If you are interested in state constrained I would suggest to refer to open-source QP solvers such as the one published here

http://ifatwww.et.uni-magdeburg.de/syst/muAO-MPC/

Or you consider: qpOASES, FORCES, ...

I hope this helps.

Conference Paper Implementation Aspects of Model Predictive Control for Embed...

Thanks Timm,

Do you have any information about implementing large LP (coming from MPC) on ARM CPUs?

No, not directly. I would check numerical methods for LPs. I never tried to solve LPs on an ARM.

OK, thanks