Hi @Mounira
This is often a problem where designers begin with what they are trying to end with. Similar to some LQR practitioners, some choose MPC because it can autotune, hoping to achieve the performance objectives by simply specifying key parameters like the prediction horizon, control horizon, sampling time, and cost function weights, thereby avoiding the extensive mathematical intervention required for manual tuning of standard feedback controllers.
However, when performance objectives aren't met, designers often find themselves tuning more parameters than the original number of control gains in standard feedback controllers. Generally, there are no hard and fast rules in tuning, but I tend to call this the "circular tuning fallacy."
Hope these articles are helpful:
