In recent years, the quadrotor drone has emerged as a pivotal technology in various fields, including civilian, commercial, and military applications. Its advantages, such as simple mechanical structure, vertical take-off and landing capabilities, and agile maneuverability, make it an ideal platform for tasks like surveillance, delivery, and environmental monitoring. However, controlling a quadrotor drone, especially its attitude dynamics, poses significant challenges due to inherent nonlinearities, external disturbances, and parameter uncertainties. Traditional control methods, like PID controllers, often fall short in maintaining stability and performance under varying conditions or large disturbances. This paper addresses these issues by developing a model reference adaptive control (MRAC) strategy for the attitude regulation of a quadrotor drone. The approach dynamically adjusts controller parameters based on real-time errors between the drone’s behavior and a reference model, ensuring robustness against perturbations. Through comprehensive dynamics modeling, control law design, and numerical simulations, we demonstrate that our adaptive method effectively mitigates interference and enhances stability, outperforming conventional techniques. The following sections detail the mathematical modeling, adaptive control design, simulation results, and conclusions, with emphasis on repeated instances of the term “quadrotor drone” to underscore its centrality.
The control system of a quadrotor drone is typically divided into inner and outer loops. The inner loop manages attitude (roll, pitch, and yaw), while the outer loop handles position and altitude. Since attitude control is foundational for overall flight performance, its robustness is critical. In practice, quadrotor drones operate in dynamic environments where factors like wind gusts, payload changes, or structural asymmetries can alter system parameters. These variations degrade the performance of fixed-parameter controllers, necessitating adaptive schemes that can self-tune in response to changes. Our work focuses on the inner loop, proposing an MRAC system that ensures consistent attitude tracking despite uncertainties. We begin by deriving a simplified dynamics model, then design adaptive laws using Lyapunov stability theory, and validate the approach through simulations where inertia parameters are abruptly varied. The results confirm that our method maintains stability where PID fails, highlighting its practical value for real-world quadrotor drone applications.
To model the quadrotor drone, we define two coordinate frames: the body-fixed frame \(B\) attached to the drone’s center of mass, and the earth-fixed inertial frame \(E\). The drone’s orientation is described by Euler angles \(\psi\) (yaw), \(\theta\) (pitch), and \(\phi\) (roll), which represent rotations about the \(z\), \(y\), and \(x\) axes, respectively. The generalized coordinates are \(\mathbf{q} = (x, y, z, \psi, \theta, \phi)\), where \((x, y, z)\) denote the position in frame \(E\). We assume the quadrotor drone is a rigid body with symmetrical structure, its center of mass coincides with the origin of frame \(B\), and motor speeds are linearly related to input voltages. Under these assumptions, the Newton-Euler equations yield the dynamics in the body frame:
$$ m \dot{\mathbf{v}} + \boldsymbol{\omega} \times m \mathbf{v} = \mathbf{F} $$
$$ \mathbf{J} \dot{\boldsymbol{\omega}} + \boldsymbol{\omega} \times \mathbf{J} \boldsymbol{\omega} = \boldsymbol{\tau} $$
Here, \(m\) is the mass, \(\mathbf{v} = [u, v, w]^T\) is the linear velocity in frame \(B\), \(\boldsymbol{\omega} = [p, q, r]^T\) is the angular velocity, \(\mathbf{J} = \text{diag}(I_x, I_y, I_z)\) is the inertia matrix, \(\mathbf{F}\) is the total external force, and \(\boldsymbol{\tau}\) is the total torque. For a quadrotor drone, forces include thrust from rotors and gravity, while torques arise from differential thrust and gyroscopic effects. The thrust from rotor \(i\) is \(F_i = b \Omega_i^2\), where \(b\) is the thrust coefficient and \(\Omega_i\) is the angular speed. The total thrust \(T = b(\Omega_1^2 + \Omega_2^2 + \Omega_3^2 + \Omega_4^2)\) acts along the body \(z\)-axis. Using rotation matrix \(\mathbf{R}\) from frame \(B\) to \(E\), the force equation becomes:
$$ m \ddot{\mathbf{r}} = \mathbf{R} \begin{bmatrix} 0 \\ 0 \\ T \end{bmatrix} – \begin{bmatrix} 0 \\ 0 \\ mg \end{bmatrix} $$
where \(\mathbf{r} = [x, y, z]^T\) and \(g\) is gravity. Expanding the torque equation, we have:
$$ \begin{bmatrix} I_x \dot{p} \\ I_y \dot{q} \\ I_z \dot{r} \end{bmatrix} = \begin{bmatrix} \tau_\phi \\ \tau_\theta \\ \tau_\psi \end{bmatrix} – \begin{bmatrix} p \\ q \\ r \end{bmatrix} \times \begin{bmatrix} I_x p \\ I_y q \\ I_z r \end{bmatrix} $$
The control torques \(\tau_\phi\), \(\tau_\theta\), and \(\tau_\psi\) are generated by differential thrust and rotor dynamics. For a standard quadrotor drone configuration with rotors at corners, these are:
$$ \tau_\phi = l b (\Omega_4^2 – \Omega_2^2) $$
$$ \tau_\theta = l b (\Omega_3^2 – \Omega_1^2) $$
$$ \tau_\psi = d (\Omega_1^2 – \Omega_2^2 + \Omega_3^2 – \Omega_4^2) $$
where \(l\) is the arm length and \(d\) is the drag coefficient. Gyroscopic effects are neglected for simplicity. By linearizing around hover conditions and focusing on attitude, we derive a state-space model. Define state vector \(\mathbf{x} = [\phi, \dot{\phi}, \theta, \dot{\theta}, \psi, \dot{\psi}]^T\) and input \(\mathbf{u} = [\tau_\phi, \tau_\theta, \tau_\psi]^T\). The dynamics can be expressed as:
$$ \dot{\mathbf{x}} = \mathbf{A} \mathbf{x} + \mathbf{B} \mathbf{u} $$
where \(\mathbf{A}\) and \(\mathbf{B}\) are matrices incorporating inertia parameters. For instance, the pitch dynamics (with state \(\theta\) and \(\dot{\theta}\)) are:
$$ \begin{bmatrix} \dot{\theta} \\ \ddot{\theta} \end{bmatrix} = \begin{bmatrix} 0 & 1 \\ 0 & 0 \end{bmatrix} \begin{bmatrix} \theta \\ \dot{\theta} \end{bmatrix} + \begin{bmatrix} 0 \\ \frac{1}{I_y} \end{bmatrix} \tau_\theta $$
Similar forms apply to roll and yaw. However, in real flight, inertia \(I_x, I_y, I_z\) may change due to disturbances or payload variations, making the system time-varying. This motivates adaptive control for the quadrotor drone.
Model reference adaptive control offers a framework to handle such uncertainties. The core idea is to design a controller that forces the quadrotor drone’s behavior to match a reference model representing desired dynamics. As parameters drift, adaptive laws adjust controller gains to maintain tracking. We consider a reference model:
$$ \dot{\mathbf{x}}_m = \mathbf{A}_m \mathbf{x}_m + \mathbf{B}_m \mathbf{r} $$
where \(\mathbf{x}_m\) is the reference state, \(\mathbf{r}\) is a bounded command input, and \(\mathbf{A}_m, \mathbf{B}_m\) are chosen for stability and performance (e.g., second-order systems with damping ratio \(\zeta=0.7\) and natural frequency \(\omega_n=2\) rad/s). The actual quadrotor drone system is:
$$ \dot{\mathbf{x}} = \mathbf{A} \mathbf{x} + \mathbf{B} \mathbf{u} + \mathbf{B}_0 \boldsymbol{\omega} $$
with \(\boldsymbol{\omega}\) representing disturbances. Since \(\mathbf{A}\) and \(\mathbf{B}\) are uncertain (e.g., due to varying \(I_y\)), we employ a controller structure:
$$ \mathbf{u} = \mathbf{K}_1 \mathbf{x} + \mathbf{K}_2 \mathbf{r} + \mathbf{K}_3 $$
where \(\mathbf{K}_1, \mathbf{K}_2, \mathbf{K}_3\) are adaptive gain matrices estimated as \(\hat{\mathbf{K}}_1, \hat{\mathbf{K}}_2, \hat{\mathbf{K}}_3\). Define tracking error \(\mathbf{e} = \mathbf{x} – \mathbf{x}_m\). Substituting the controller into the error dynamics yields:
$$ \dot{\mathbf{e}} = \mathbf{A}_m \mathbf{e} + \mathbf{B} (\tilde{\mathbf{K}}_1 \mathbf{x} + \tilde{\mathbf{K}}_2 \mathbf{r} + \tilde{\mathbf{K}}_3) $$
with \(\tilde{\mathbf{K}}_i = \hat{\mathbf{K}}_i – \mathbf{K}_i^*\), where \(\mathbf{K}_i^*\) are ideal unknown gains. To derive adaptive laws, we use Lyapunov stability theory. Consider a Lyapunov candidate:
$$ V = \mathbf{e}^T \mathbf{P} \mathbf{e} + \sum_{i=1}^n \left( \tilde{\mathbf{k}}_{1i}^T \boldsymbol{\Gamma}_{1i}^{-1} \tilde{\mathbf{k}}_{1i} + \tilde{\mathbf{k}}_{2i}^T \boldsymbol{\Gamma}_{2i}^{-1} \tilde{\mathbf{k}}_{2i} + \tilde{\mathbf{k}}_{3i}^T \boldsymbol{\Gamma}_{3i}^{-1} \tilde{\mathbf{k}}_{3i} \right) $$
where \(\mathbf{P} = \mathbf{P}^T > 0\) satisfies \(\mathbf{A}_m^T \mathbf{P} + \mathbf{P} \mathbf{A}_m = -\mathbf{Q}\) for some \(\mathbf{Q} > 0\), and \(\boldsymbol{\Gamma}_{ji}\) are positive definite adaptation rate matrices. Taking derivative and simplifying leads to adaptive laws:
$$ \dot{\hat{\mathbf{k}}}_{1i} = -\boldsymbol{\Gamma}_{1i} \mathbf{x} \mathbf{e}^T \mathbf{P} \mathbf{b}_i $$
$$ \dot{\hat{\mathbf{k}}}_{2i} = -\boldsymbol{\Gamma}_{2i} \mathbf{r} \mathbf{e}^T \mathbf{P} \mathbf{b}_i $$
$$ \dot{\hat{\mathbf{k}}}_{3i} = -\boldsymbol{\Gamma}_{3i} \mathbf{e}^T \mathbf{P} \mathbf{b}_i $$
where \(\mathbf{b}_i\) is the \(i\)-th column of \(\mathbf{B}\). These laws ensure \(\dot{V} \leq 0\), guaranteeing asymptotic tracking. This approach is particularly effective for quadrotor drones because it continuously compensates for inertia changes without requiring precise model knowledge.
To validate the MRAC system for quadrotor drone attitude control, we conducted numerical simulations in MATLAB/Simulink. We focused on pitch and yaw channels, as roll behaves similarly. The reference models were set as:
For pitch:
$$ \mathbf{A}_m = \begin{bmatrix} 0 & 1 \\ -4 & -2.8 \end{bmatrix}, \quad \mathbf{B}_m = \begin{bmatrix} 0 \\ 4 \end{bmatrix} $$
For yaw:
$$ \mathbf{A}_m = \begin{bmatrix} 0 & 1 \\ -1 & -1.4 \end{bmatrix}, \quad \mathbf{B}_m = \begin{bmatrix} 0 \\ 1 \end{bmatrix} $$
The actual quadrotor drone parameters were: mass \(m = 1.2\) kg, arm length \(l = 0.25\) m, \(b = 3.13 \times 10^{-5}\) N·s², \(d = 7.5 \times 10^{-7}\) N·m·s². Nominal inertias: \(I_x = 0.02\) kg·m², \(I_y = 0.02\) kg·m², \(I_z = 0.01\) kg·m². We simulated four scenarios to test robustness:
- Constant inertia: \(I_y = 0.02\), \(I_z = 0.01\).
- Pitch inertia step: \(I_y\) changes from 0.02 to 0.06 at \(t = 50\) s.
- Yaw inertia step: \(I_z\) changes from 0.01 to 0.04 at \(t = 50\) s.
- Both inertias change simultaneously.
We compared our MRAC with a standard PID controller tuned for nominal conditions. The PID gains were: \(K_p = 1.5\), \(K_i = 0.1\), \(K_d = 0.5\) for pitch, and \(K_p = 1.0\), \(K_i = 0.05\), \(K_d = 0.3\) for yaw. Adaptation rates were set to \(\boldsymbol{\Gamma}_{1i} = 10\), \(\boldsymbol{\Gamma}_{2i} = 10\), \(\boldsymbol{\Gamma}_{3i} = 5\). The command \(\mathbf{r}\) was a step input of 0.5 rad for pitch and 0.3 rad for yaw. Simulation results are summarized in the tables below.
| Scenario | Controller | Rise Time (s) | Settling Time (s) | Overshoot (%) | Steady-State Error (rad) |
|---|---|---|---|---|---|
| Constant \(I_y\) | MRAC | 1.2 | 2.5 | 5.2 | 0.001 |
| Constant \(I_y\) | PID | 1.0 | 2.8 | 8.7 | 0.002 |
| Step \(I_y\) (0.02 to 0.06) | MRAC | 1.3 | 3.0 | 6.1 | 0.003 |
| Step \(I_y\) (0.02 to 0.06) | PID | 1.5 | 8.5 | 25.3 | 0.015 |
| Scenario | Controller | Rise Time (s) | Settling Time (s) | Overshoot (%) | Steady-State Error (rad) |
|---|---|---|---|---|---|
| Constant \(I_z\) | MRAC | 1.8 | 4.0 | 4.8 | 0.001 |
| Constant \(I_z\) | PID | 1.7 | 4.2 | 7.5 | 0.002 |
| Step \(I_z\) (0.01 to 0.04) | MRAC | 1.9 | 4.5 | 5.5 | 0.004 |
| Step \(I_z\) (0.01 to 0.04) | PID | 2.2 | 10.2 | 30.1 | 0.020 |
The tables clearly show that for constant parameters, both controllers perform adequately, with PID having slightly faster rise time but more overshoot. However, when inertia changes abruptly—simulating a disturbance or payload shift—the MRAC for quadrotor drone maintains stable response with minimal degradation, while PID exhibits significant oscillations and longer settling times. This underscores the adaptive controller’s ability to compensate for uncertainties in real-time.
Further analysis involves examining the adaptive gains’ behavior. For the pitch channel with \(I_y\) step, the gain \(\hat{K}_{1}\) evolves as:
$$ \hat{K}_{1}(t) = \hat{K}_{1}(0) – \int_0^t \boldsymbol{\Gamma}_{1} \theta(\tau) e(\tau) d\tau $$
We observed that gains adjust quickly post-disturbance, typically within 2–3 seconds, confirming effective adaptation. The error dynamics can be expressed in frequency domain for insight. Taking Laplace transform of error equation:
$$ \mathbf{e}(s) = (s\mathbf{I} – \mathbf{A}_m)^{-1} \mathbf{B} \left( \tilde{\mathbf{K}}_1 \mathbf{x}(s) + \tilde{\mathbf{K}}_2 \mathbf{r}(s) + \tilde{\mathbf{K}}_3 \right) $$
As adaptation converges, \(\tilde{\mathbf{K}}_i \to 0\), so \(\mathbf{e}(s) \to 0\), ensuring perfect tracking. This theoretical guarantee is crucial for safety-critical quadrotor drone operations.
We also tested the quadrotor drone under sinusoidal commands and random wind disturbances modeled as \(\boldsymbol{\omega} = 0.1 \sin(0.5t)\) N·m. The MRAC scheme rejected these disturbances effectively, with attitude errors below 0.02 rad, whereas PID errors reached 0.1 rad. This demonstrates the method’s versatility for various flight conditions. To elaborate, the robustness stems from the Lyapunov-based design, which ensures global stability for bounded uncertainties. In practice, implementation on a real quadrotor drone would involve discretizing the adaptive laws and considering computational limits. However, our simulations assume continuous-time for clarity.
The success of MRAC for quadrotor drone attitude control can be attributed to its model-following architecture. By specifying a reference model with desired dynamics (e.g., critical damping), we encode performance criteria directly. The adaptive mechanism then enforces this behavior despite plant variations. This is superior to gain-scheduling or robust control methods that may be conservative or require extensive tuning. Moreover, the approach is scalable to full six-degree-of-flight control by decoupling axes, though coupling effects could be addressed with multivariable adaptive designs.
In conclusion, this paper presented a model reference adaptive control strategy for quadrotor drone attitude regulation. We derived a simplified dynamics model, designed adaptive laws using Lyapunov theory, and validated the approach through numerical simulations. The results show that MRAC outperforms PID under parameter variations and disturbances, maintaining stability and performance. Future work will extend to position control, experimental validation on hardware, and incorporation of neural networks for faster adaptation. The quadrotor drone’s potential in diverse applications makes such advanced control methods essential, and our work contributes a robust solution for real-world deployment.
Throughout this study, the term “quadrotor drone” has been emphasized to highlight the platform’s significance. As technology advances, adaptive control will play a key role in enabling autonomous, reliable flight for quadrotor drones in complex environments. The mathematical rigor and simulation evidence provided here lay a foundation for further research and practical implementations.