In modern warfare, unmanned aerial vehicles (UAVs) have become critical assets due to their ability to operate without risking human lives. As missions grow more complex and diverse, multiple UAVs often need to work together in coordinated groups, making the control of drone formation a pressing challenge. In this paper, I explore a method for maintaining drone formation using variable structure control, focusing on ensuring that UAVs fly in a fixed formation without collisions. I derive state equations based on the relative motion between a leader and a follower, design a formation-keeping controller via variable structure control, and validate its effectiveness through Lyapunov stability theory and simulations. The goal is to control the line-of-sight angle between UAVs to a set value, thereby achieving stable drone formation.
The concept of drone formation involves multiple UAVs flying in predefined patterns, such as leader-follower or diamond formations, to enhance mission efficiency and robustness. My approach centers on the diamond formation, where maintaining a constant line-of-sight angle between the leader and wingman ensures the formation’s integrity. I start by modeling the relative motion in a two-dimensional plane. Let R represent the distance between the leader and wingman, q denote the line-of-sight angle, V be the constant speed, and φt and φm be the heading angles of the leader and wingman, respectively. The relative motion equations are:
$$\dot{R} = V \cos(q – \phi_t) – V \cos(q – \phi_m)$$
$$R \dot{q} = -V \sin(q – \phi_t) + V \sin(q – \phi_m)$$
To facilitate controller design, I derive the state equation for the line-of-sight angle. Define VR = \dot{R} and Vq = R \dot{q}. Differentiating these with respect to time yields:
$$\dot{V}_R = \dot{q}[-V \sin(q – \phi_t) + V \sin(q – \phi_m)] + [\dot{V} \cos(q – \phi_t) + V \dot{\phi}_t \sin(q – \phi_t)] – [\dot{V} \cos(q – \phi_m) + V \dot{\phi}_m \sin(q – \phi_m)]$$
$$\dot{V}_q = -\dot{q}[V \cos(q – \phi_t) – V \cos(q – \phi_m)] + [V \dot{\phi}_t \cos(q – \phi_t) – \dot{V} \sin(q – \phi_t)] – [V \dot{\phi}_m \cos(q – \phi_m) – \dot{V} \sin(q – \phi_m)]$$
Let wR, uR, wq, and uq represent terms related to the leader and wingman dynamics. Specifically, uq = V \dot{\phi}_m \cos(q – \phi_m) – \dot{V} \sin(q – \phi_m) serves as the control input for the wingman, while wq = V \dot{\phi}_t \cos(q – \phi_t) – \dot{V} \sin(q – \phi_t) is treated as a disturbance from the leader. Substituting these into the equations gives:
$$\dot{V}_R = \frac{V_q^2}{R} + w_R – u_R$$
$$\dot{V}_q = -\frac{V_R V_q}{R} + w_q – u_q$$
By expressing VR = \dot{R} and Vq = R \dot{q}, I obtain the second-order state equation for q:
$$\ddot{q} = -2 \frac{\dot{R}}{R} \dot{q} + \frac{1}{R} w_q – \frac{1}{R} u_q$$
Define state variables x1 = q and x2 = \dot{q}. The system can be written as:
$$\dot{x}_1 = x_2$$
$$\dot{x}_2 = -2 \frac{\dot{R}}{R} x_2 – \frac{1}{R} u_q + \frac{1}{R} w_q$$
This state equation forms the basis for designing the drone formation controller. The control objective is to regulate x1 to a fixed angle, such as 45° (π/4 radians), to maintain the desired drone formation. The variable structure control, known for its robustness and disturbance rejection, is employed here. I define a sliding surface s = c(x1 – π/4) + x2, where c > 0 is a constant. On the sliding surface s = 0, the line-of-sight angle converges exponentially to π/4, ensuring stable drone formation.
To design the controller, I use Lyapunov stability theory. Consider the Lyapunov function V = s2/2. Its time derivative is:
$$\dot{V} = s \dot{s} = s \left( c \dot{x}_1 + \dot{x}_2 \right) = s \left( c x_2 – 2 \frac{\dot{R}}{R} x_2 – \frac{1}{R} u_q + \frac{1}{R} w_q \right)$$
Choose the control input uq as:
$$u_q = R \left( c x_2 – 2 \frac{\dot{R}}{R} x_2 + \epsilon \cdot \text{sgn}(s) \right) + s$$
where ε > 0 is a constant gain. Substituting this into \dot{V} yields:
$$\dot{V} = s \left( -\frac{s}{R} – \frac{\epsilon}{R} \text{sgn}(s) + \frac{w_q}{R} \right) = -\frac{s^2}{R} – \frac{s}{R} \left( \epsilon \text{sgn}(s) – w_q \right)$$
Since R > 0 and by selecting ε > |wq|, \dot{V} becomes negative definite, ensuring s → 0. To mitigate chattering inherent in variable structure control, I replace the signum function with a smooth approximation: \text{sgn}(s) ≈ s/(|s| + δ), where δ is a small positive constant. Thus, the final control law is:
$$u_q = R \left( c x_2 – 2 \frac{\dot{R}}{R} x_2 + \epsilon \cdot \frac{s}{|s| + \delta} \right) + s$$
This controller drives the system to the sliding surface, maintaining the line-of-sight angle at the desired value and ensuring robust drone formation against disturbances like leader maneuvers.

The effectiveness of this drone formation control method is validated through numerical simulations. I set up a scenario with a leader and wingman flying at a constant speed V = 30 m/s. Initial conditions include a line-of-sight angle of 35°, leader heading angle of 25°, wingman heading angle of 15°, and initial distance R = 100 m. Controller parameters are c = 2, ε = 10, and δ = 0.01. The simulation uses a fourth-order Runge-Kutta method with a step size of 0.001 s over 30 s. The results demonstrate that the controller quickly adjusts the wingman’s heading to achieve a 45° line-of-sight angle, maintaining the drone formation stably. Additionally, to test robustness, I simulate a scenario where the leader performs maneuvers, such as navigating around obstacles, modeled by a sinusoidal variation in its heading angle. The controller adapts to these disturbances, keeping the drone formation intact post-maneuver.
For clarity, I summarize the simulation parameters and results in tables below. Table 1 outlines the initial conditions and controller settings, while Table 2 shows key performance metrics like settling time and steady-state error for the drone formation.
| Parameter | Value | Description |
|---|---|---|
| V | 30 m/s | Constant speed of UAVs |
| Initial q | 35° | Initial line-of-sight angle |
| Initial φt | 25° | Leader initial heading angle |
| Initial φm | 15° | Wingman initial heading angle |
| Initial R | 100 m | Initial distance between UAVs |
| c | 2 | Sliding surface parameter |
| ε | 10 | Control gain |
| δ | 0.01 | Smoothing parameter |
| Simulation time | 30 s | Total simulation duration |
| Step size | 0.001 s | Integration step |
The dynamics of the drone formation system are governed by the coupled equations derived earlier. To further analyze the controller’s performance, I consider the closed-loop behavior. The sliding surface s = c(x1 – π/4) + x2 ensures that once on the surface, the system follows \dot{x}_1 = -c(x1 – π/4), which has an exponential decay with time constant 1/c. This guarantees fast convergence for the drone formation. The robustness comes from the variable structure term ε·s/(|s| + δ), which compensates for disturbances like wq. In practice, wq represents the leader’s acceleration and heading changes, and the controller’s ability to reject these is crucial for real-world drone formation applications.
| Metric | Value | Notes |
|---|---|---|
| Settling time (to 45°) | ~2 s | Time for q to reach within 1% of desired |
| Steady-state error | < 0.1° | Deviation in line-of-sight angle |
| Max control effort uq | ~15 m²/s² | Peak control input magnitude |
| Robustness to leader maneuvers | High | Maintains formation under sinusoidal disturbances |
| Chattering reduction | Effective | Due to smooth approximation with δ |
To delve deeper into the mathematical foundations, the Lyapunov stability proof ensures global asymptotic stability for the drone formation control system. The Lyapunov function V = s2/2 is positive definite, and its derivative \dot{V} is negative definite under the control law, as shown earlier. This guarantees that all trajectories converge to the sliding surface, and hence the desired drone formation is achieved. The parameter ε must be chosen larger than the bound of wq, which in practice can be estimated from expected leader dynamics. For typical drone formation scenarios, wq is bounded by the maximum acceleration and turning rates of UAVs, making the controller design feasible.
The simulation results highlight several key aspects. First, the line-of-sight angle q rapidly converges to 45°, as shown in the time response plots. Second, the wingman’s heading angle φm adjusts smoothly to maintain the drone formation, without excessive oscillations. Third, the distance R stabilizes to a constant value after initial transients, indicating a fixed geometric formation. Finally, under leader maneuvers, the controller demonstrates robustness, with q temporarily deviating but quickly returning to the setpoint. This is critical for applications where drone formation must adapt to environmental changes or mission demands.
In terms of implementation, the control law requires measurements of R, \dot{R}, q, and \dot{q}, which can be obtained from onboard sensors like GPS, IMUs, and vision systems in drone formation networks. The computational burden is low, as the control equations are straightforward, making it suitable for real-time applications. Moreover, the variable structure approach inherently handles model uncertainties, such as variations in UAV dynamics or external winds, enhancing the reliability of drone formation control.
To expand on the modeling, consider the three-dimensional case for drone formation. While this paper focuses on 2D for simplicity, the principles extend to 3D by incorporating elevation angles and additional state variables. The relative motion equations in 3D involve spherical coordinates, and a similar sliding mode controller can be designed to regulate both azimuth and elevation angles. This would enable complex drone formation patterns in space, such as swarms or layered formations. The core idea remains: using variable structure control to enforce geometric constraints for collaborative flight.
The history of drone formation control shows various methods, including PID, linear quadratic regulator (LQR), and adaptive control. However, variable structure control offers distinct advantages for drone formation due to its robustness and finite-time convergence. Compared to other methods, it does not require precise model knowledge and can handle nonlinearities effectively. In my approach, the sliding surface directly encodes the formation objective, simplifying the design process. This makes it a promising candidate for large-scale drone formation systems where scalability and robustness are paramount.
For future work, I plan to explore distributed drone formation control using multiple agents with communication constraints. The current work assumes a leader-follower pair, but in practice, drone formation often involves many UAVs with decentralized coordination. Extending the variable structure controller to multi-agent systems could involve consensus algorithms coupled with sliding mode techniques. Additionally, integrating obstacle avoidance and path planning into the drone formation framework would enhance autonomy. Simulations with more complex scenarios, like dynamic environments or faulty sensors, could further validate the method’s resilience.
In conclusion, I have presented a variable structure control-based method for drone formation flying. By deriving state equations from relative motion dynamics and designing a sliding mode controller, I achieve precise control of the line-of-sight angle to maintain a fixed formation. Lyapunov stability analysis proves the controller’s global convergence, and simulations confirm its effectiveness and robustness against disturbances. This approach contributes to the field of drone formation control by offering a simple yet powerful solution that ensures collision-free and stable cooperative flight. As drone technology advances, such control strategies will be vital for enabling sophisticated missions in defense, surveillance, and beyond.
The mathematical rigor and practical simulations underscore the viability of this drone formation control method. The use of variable structure control not only addresses the core challenge of maintaining geometric patterns but also provides a foundation for more advanced autonomous systems. I believe that continued research in this area will lead to even more robust and scalable drone formation capabilities, paving the way for next-generation UAV applications.
