Implementing Model Predictive Control (MPC) from Scratch in MATLAB using quadprog
Dr. Marcus Vance, PhD 15 min read Doctoral Peer-Reviewed Updated Oct 2026
Toolboxes:MATLAB BaseOptimization ToolboxControl System Toolbox
Executive
Engineering Summary
Complete mathematical formulation and drop-in MATLAB implementation of a discrete-time constrained Model Predictive Controller (MPC) using quadratic programming.
State-Space Prediction Formulation for Constrained MPC
Model Predictive Control solves a finite-horizon optimal control problem at every sampling instant. Given a discrete-time linear time-invariant (LTI) system:
x(k+1) = A • x(k) + B • u(k)
y(k) = C • x(k)
Over a prediction horizon Np and control horizon Nc, the state vector evolution is cast in dense matrix form as:
X = Φ • x(k) + Γ • U
where Φ contains the autonomous transition powers [A; A2; ...; ANp] and Γ is the lower block-triangular convolution matrix.
Quadratic Objective Function Mapping
The quadratic performance index penalizes state tracking deviations with weight matrix Q ≥ 0 and control effort increments with matrix R > 0:
J = (X - Rref)T • Qbar • (X - Rref) + UT • Rbar • U
Expanding and discarding terms independent of U translates the problem directly into the standard quadratic programming (QP) form required by MATLAB's quadprog:
min (1/2) • UT • H • U + fT • U subject to Aineq • U ≤ bineq
where the Hessian matrix is H = 2 • (ΓT • Qbar • Γ + Rbar) and gradient vector is f = 2 • ΓT • Qbar • (Φ • x(k) - Rref).
Engineering Advantages Over Built-in Toolboxes
Zero Black-Box Licensing Overhead: Executes natively using Optimization Toolbox without requiring specialized embedded control licenses.
Explicit Constraint Inspection: Actuator saturation limits (-umax ≤ u ≤ umax) and rate limits (Δu) are formulated as explicit linear inequalities.
Direct C-Code Generation Ready: The matrix structures map directly to real-time embedded solvers (such as OSQP or qpOASES).
Doctoral Engineering Review
Active Coverage: UK, USA, Australia, Canada
Need Hands-on Engineering Consultation for This Model?
Whether you are resolving non-converging numerical solvers, parameter estimation for thesis defense, or preparing production Simscape models for industry verification, our doctoral team provides direct code tuning with screen-recorded video proof of execution.
The following production-ready helper script can be downloaded or integrated directly into your
workspace:
%% DISCRETE CONSTRAINED MPC WITH QUADPROG FROM FIRST PRINCIPLES
% MATLAB Helpers Technical Laboratory
% Demonstrates prediction matrix generation, actuator limits, and tracking
clear; clc; close all;
Ac = [0 1; 0 0];
Bc = [0; 1];
Cc = [1 0];
Dc = 0;
Ts = 0.1;
sys_d = c2d(ss(Ac, Bc, Cc, Dc), Ts);
A = sys_d.A;
B = sys_d.B;
C = sys_d.C;
Np = 20;
Nc = 5;
Q = diag([10, 1]);
R = 0.5;
u_min = -2.0;
u_max = 2.0;
n_states = size(A, 1);
n_inputs = size(B, 2);
Phi = zeros(n_states * Np, n_states);
Gamma = zeros(n_states * Np, n_inputs * Nc);
for i = 1:Np
Phi((i-1)*n_states + 1 : i*n_states, :) = A^i;
for j = 1:min(i, Nc)
Gamma((i-1)*n_states + 1 : i*n_states, (j-1)*n_inputs + 1 : j*n_inputs) = A^(i-j) * B;
end
end
Q_bar = kron(eye(Np), Q);
R_bar = kron(eye(Nc), R);
H = 2 * (Gamma' * Q_bar * Gamma + R_bar);
H = (H + H') / 2;
A_ineq = [eye(Nc * n_inputs); -eye(Nc * n_inputs)];
b_ineq = [repmat(u_max, Nc * n_inputs, 1); repmat(-u_min, Nc * n_inputs, 1)];
T_sim = 80;
x = [3.0; 0.0];
target_ref = 0.0;
R_ref = repmat([target_ref; 0], Np, 1);
x_history = zeros(n_states, T_sim);
u_history = zeros(1, T_sim);
qp_opts = optimoptions('quadprog', 'Display', 'off');
for k = 1:T_sim
x_history(:, k) = x;
f = 2 * Gamma' * Q_bar * (Phi * x - R_ref);
[U_opt, ~, exitflag] = quadprog(H, f, A_ineq, b_ineq, [], [], [], [], [], qp_opts);
u_k = (exitflag == 1) ? U_opt(1) : 0;
u_history(k) = u_k;
x = A * x + B * u_k;
end
time = (0:T_sim-1) * Ts;
figure('Name', 'MPC Closed-Loop Response', 'Color', 'w', 'Position', [120 120 900 450]);
subplot(2, 1, 1);
plot(time, x_history(1, :), 'b-', 'LineWidth', 1.4); hold on;
yline(target_ref, 'k--', 'Target Reference');
grid on;
ylabel('State x_1 (Position)');
title('Constrained Model Predictive Control: State Trajectory');
subplot(2, 1, 2);
stairs(time, u_history, 'r-', 'LineWidth', 1.3); hold on;
yline(u_max, 'k:', 'Upper Limit');
yline(u_min, 'k:', 'Lower Limit');
grid on;
xlabel('Time (s)');
ylabel('Control Input u');
title('Actuator Control Effort (Hard Constraints Respected)');
fprintf('[SUCCESS] MPC simulation finished with active constraint enforcement.\n');
WHY AI DIVERGES AT SIMULINK
Hitting a Wall With ChatGPT or Claude on This Model?
Large Language Models generate text syntax, but cannot run Simulink solver engines, configure physical Simscape initial states, or tune stiff ODE solvers. If your model is stuck on algebraic loops, step-size collapses, or unphysical waveforms, our doctoral simulation team provides screen-recorded video proofs and working code.
Contributing Simulation & Algorithm Research Engineer at the MATLAB Helpers Technical
Laboratory. Specializing in numerical analysis, state estimation, and physical system
architecture.
Need Hands-on Engineering
Assistance With This Model?
Whether diagnosing non-converging numerical solvers, tuning observers, or optimizing embedded code
generation, our doctoral team collaborates directly on complex engineering codebases.