Model predictive control

Model predictive control (MPC) is an advanced control strategy that uses a dynamic model of the system to predict future behavior over a finite horizon, solves an optimization problem at each time step to determine the optimal control sequence, and applies only the first control action before repeating the process. Unlike classical feedback controllers that react to current error, MPC anticipates future constraints and disturbances, making it the dominant paradigm in process industries, robotics, and autonomous vehicles where constraints are as important as performance. The method's power comes from its explicit handling of multivariable interactions and hard constraints, but its fragility comes from the same source: the optimizer's solution is only as good as the model's predictions, and the computational delay between measurement and actuation creates a hidden feedback loop that can destabilize the system when the horizon is too short or the model is wrong.

MPC occupies a unique position in the control landscape: it is formally a feedforward control strategy because it predicts the future, but operationally it is a feedback strategy because it recomputes at every step. This duality is not merely a classification puzzle — it reveals that the distinction between feedforward and feedback breaks down when the controller's internal model is sophisticated enough to include itself. The controller that models its own future actions is a system that has swallowed its own feedback loop, and the consequences of this recursion are still not fully understood.