Precision Positioning and Swing Suppression for Quadrotor Drone with Double-Pendulum Payload

In recent years, the quadrotor drone has evolved from a platform primarily used for aerial photography and reconnaissance to a versatile tool for aerial manipulation and transportation. Among its many applications, suspended load transportation offers significant advantages over rigidly attached carriage, as it is less constrained by the payload’s geometry and reduces the impact on the quadrotor drone‘s moment of inertia, thereby enhancing maneuverability. However, this method introduces a critical challenge: the induction of undesirable pendulum-like oscillations in the suspended cargo. While substantial research exists on controlling single-pendulum systems in quadrotor drone lift operations, practical scenarios often involve a non-negligible hook mass or a large payload that cannot be treated as a point mass. This gives rise to a double-pendulum effect, where both the hook and the payload swing, presenting a more complex, underactuated control problem. This phenomenon, if unaddressed, can compromise operational safety, positioning accuracy, and system stability.

The dynamics of a quadrotor drone carrying a double-pendulum payload are significantly more intricate than its single-pendulum counterpart. The system exhibits two inherent natural frequencies of oscillation, leading to more persistent and complex swing modes. Control strategies for such systems can broadly be classified into closed-loop and open-loop methods. Closed-loop approaches typically rely on sensors to measure states like swing angles and incorporate this feedback into nonlinear controllers, such as those based on energy shaping or backstepping. While effective, these methods often increase system complexity and cost due to the required sensing apparatus. Open-loop methods, including input shaping and trajectory planning, offer an alternative by pre-designing the quadrotor drone‘s motion profile to naturally dampen oscillations without direct swing angle feedback. Online trajectory planning, in particular, is promising as it can adapt the planned path in real-time based on system dynamics or estimated states, blending the robustness of feedback with the simplicity of open-loop command generation.

This article presents a comprehensive methodology for the precise positioning and swing suppression of a quadrotor drone subject to double-pendulum dynamics. The core of the approach is an online trajectory planning framework that synthesizes a reference positioning path with an actively calculated anti-swing component. The proposed control law ensures asymptotic convergence of both the quadrotor drone to its target destination and the swing angles of the hook and payload to zero. A detailed mathematical model is derived, the stability and convergence properties are rigorously proven using Lyapunov and Barbalat’s tools, and the effectiveness of the method is validated through numerical simulations.

Mathematical Modeling of the Double-Pendulum Quadrotor Drone System

Accurate modeling is the foundation for effective controller design. We consider a two-dimensional analysis where the quadrotor drone motion is confined to the vertical plane. This simplification is common for analyzing translational dynamics and pendulum effects, with the understanding that control in the perpendicular plane would follow an analogous structure. The system comprises three primary masses: the quadrotor drone of mass \(M\), the hook of mass \(m_1\), and the payload of mass \(m_2\). They are connected via two massless, inextensible cables of lengths \(l_1\) (drone to hook) and \(l_2\) (hook to payload). The following key assumptions are made to derive the model:

  1. The quadrotor drone body is rigid.
  2. Gravitational acceleration \(g\) is constant.
  3. All connection points (drone-cable, cable-cable) are ideal spherical joints with no friction.
  4. Cable masses are negligible.

The system’s configuration is defined by the generalized coordinates \(\mathbf{q}(t) = [x(t), z(t), \theta_1(t), \theta_2(t)]^T \in \mathbb{R}^4\), where \((x, z)\) is the quadrotor drone‘s position in the inertial frame, \(\theta_1\) is the hook’s angular deviation from the vertical, and \(\theta_2\) is the payload’s angular deviation from the vertical. The positions of the hook \((x_{m_1}, z_{m_1})\) and payload \((x_{m_2}, z_{m_2})\) are kinematically related to the drone’s position:

$$
\begin{aligned}
x_{m_1} &= x + l_1 \sin\theta_1, \quad &z_{m_1} &= z – l_1 \cos\theta_1 \\
x_{m_2} &= x + l_1 \sin\theta_1 + l_2 \sin\theta_2, \quad &z_{m_2} &= z – l_1 \cos\theta_1 – l_2 \cos\theta_2
\end{aligned}
$$

The Lagrangian method is employed to derive the equations of motion. The system’s kinetic energy \(T\) and potential energy \(U\) are given by:

$$
\begin{aligned}
T &= \frac{1}{2}M(\dot{x}^2 + \dot{z}^2) + \frac{1}{2}m_1(\dot{x}_{m_1}^2 + \dot{z}_{m_1}^2) + \frac{1}{2}m_2(\dot{x}_{m_2}^2 + \dot{z}_{m_2}^2) \\
U &= (M+m_1+m_2)gz – (m_1+m_2)gl_1\cos\theta_1 – m_2 g l_2 \cos\theta_2
\end{aligned}
$$

The Lagrangian is \(L = T – U\). Applying the Euler-Lagrange equation,

$$
\frac{d}{dt}\left( \frac{\partial L}{\partial \dot{\mathbf{q}}} \right) – \frac{\partial L}{\partial \mathbf{q}} = \mathbf{F}
$$

where \(\mathbf{F} = [F_x, F_z, 0, 0]^T\) represents the generalized forces (the total thrust components on the quadrotor drone), yields the following nonlinear equations of motion:

$$
\begin{aligned}
&(M+m_1+m_2)\ddot{x} + (m_1+m_2)l_1(\ddot{\theta}_1\cos\theta_1 – \dot{\theta}_1^2\sin\theta_1) + m_2 l_2 (\ddot{\theta}_2\cos\theta_2 – \dot{\theta}_2^2\sin\theta_2) = F_x \\
&(M+m_1+m_2)\ddot{z} + (m_1+m_2)l_1(\ddot{\theta}_1\sin\theta_1 + \dot{\theta}_1^2\cos\theta_1) + m_2 l_2 (\ddot{\theta}_2\sin\theta_2 + \dot{\theta}_2^2\cos\theta_2) + (M+m_1+m_2)g = F_z \\
&(m_1+m_2)l_1(\ddot{x}\cos\theta_1 + \ddot{z}\sin\theta_1 + g\sin\theta_1) + (m_1+m_2)l_1^2\ddot{\theta}_1 + m_2 l_1 l_2 (\ddot{\theta}_2\cos(\theta_1-\theta_2) + \dot{\theta}_2^2\sin(\theta_1-\theta_2)) = 0 \\
&m_2 l_2 (\ddot{x}\cos\theta_2 + \ddot{z}\sin\theta_2 + g\sin\theta_2) + m_2 l_2^2 \ddot{\theta}_2 + m_2 l_1 l_2 (\ddot{\theta}_1\cos(\theta_1-\theta_2) – \dot{\theta}_1^2\sin(\theta_1-\theta_2)) = 0
\end{aligned}
$$

The last two equations, which contain zero on the right-hand side, describe the unforced dynamics of the double pendulum. The roll angle \(\Phi(t)\) of the quadrotor drone, required to generate the net horizontal force \(F_x\), can be computed from the force components as \(\Phi = \arctan(-F_x / F_z)\), assuming small angle approximations for attitude dynamics or a lower-level attitude controller.

Online Trajectory Planning and Control Design

The control objective is twofold: drive the quadrotor drone to a desired target position \((d_x, d_z)\) and simultaneously suppress the swing angles \(\theta_1\) and \(\theta_2\) to zero. The proposed method achieves this by planning an acceleration trajectory for the quadrotor drone that consists of two additive components:

$$
\begin{aligned}
\ddot{x}_c(t) &= \ddot{x}_1(t) + \ddot{x}_2(t) \\
\ddot{z}_c(t) &= \ddot{z}_1(t) + \ddot{z}_2(t)
\end{aligned}
$$

Here, \(\ddot{x}_1(t)\) and \(\ddot{z}_1(t)\) form the positioning trajectory designed to guide the quadrotor drone to its destination. The terms \(\ddot{x}_2(t)\) and \(\ddot{z}_2(t)\) constitute the anti-swing component, actively calculated to damp pendulum oscillations without interfering with the final positioning of the quadrotor drone.

Design of the Positioning Trajectory

The positioning trajectory must be smooth to avoid exciting high-frequency dynamics and must satisfy boundary conditions for precise positioning. A smooth S-curve profile, such as one based on hyperbolic functions, is an excellent choice. It must satisfy the following constraints:

  1. Convergence: \(\lim_{t\to\infty} x_1(t) = d_x\) and \(\lim_{t\to\infty} z_1(t) = d_z\).
  2. Smoothness and Boundedness: The velocity, acceleration, and jerk must be bounded and converge to zero: \(\lim_{t\to\infty} \dot{x}_1, \ddot{x}_1, \dddot{x}_1, \dot{z}_1, \ddot{z}_1, \dddot{z}_1 = 0\).
  3. Initial Conditions: The system starts at rest: \(x_1(0)=\dot{x}_1(0)=z_1(0)=\dot{z}_1(0)=\theta_1(0)=\dot{\theta}_1(0)=\theta_2(0)=\dot{\theta}_2(0)=0\).

A specific instance of such a trajectory is given by:

$$
\begin{aligned}
x_1(t) &= \frac{d_x}{2} + \frac{1}{2k_{2x}} \ln\left[ \frac{\cosh(k_{1x}t – \varepsilon)}{\cosh(k_{1x}t – \varepsilon – 2k_{2x}d_x)} \right] \\
z_1(t) &= \frac{d_z}{2} + \frac{1}{2k_{2z}} \ln\left[ \frac{\cosh(k_{1z}t – \varepsilon)}{\cosh(k_{1z}t – \varepsilon – 2k_{2z}d_z)} \right]
\end{aligned}
$$

where \(k_{1x}, k_{2x}, k_{1z}, k_{2z},\) and \(\varepsilon\) are positive tuning parameters that shape the velocity and acceleration profiles.

Design of the Anti-Swing Component

The anti-swing component is derived using a Lyapunov-based analysis to guarantee swing suppression. We first analyze the dynamics of the double pendulum, which can be rewritten from the last two equations of motion as:

$$
\begin{aligned}
&l_1(\ddot{x}\cos\theta_1 + \ddot{z}\sin\theta_1 + g\sin\theta_1) + l_1^2\ddot{\theta}_1 + \frac{m_2}{m_1+m_2} l_1 l_2 \left[ \ddot{\theta}_2\cos(\theta_1-\theta_2) + \dot{\theta}_2^2\sin(\theta_1-\theta_2) \right] = 0 \\
&\frac{m_2}{m_1+m_2}l_2(\ddot{x}\cos\theta_2 + \ddot{z}\sin\theta_2 + g\sin\theta_2) + \frac{m_2}{m_1+m_2}l_2^2\ddot{\theta}_2 + \frac{m_2}{m_1+m_2} l_1 l_2 \left[ \ddot{\theta}_1\cos(\theta_1-\theta_2) – \dot{\theta}_1^2\sin(\theta_1-\theta_2) \right] = 0
\end{aligned}
$$

Consider a candidate Lyapunov function \(V(t)\) representing a measure of the pendulum system’s mechanical energy (kinetic and potential relative to the lowest point):

$$
\begin{aligned}
V(t) &= \frac{1}{2} l_1^2 \dot{\theta}_1^2 + \frac{1}{2}\frac{m_2}{m_1+m_2} l_2^2 \dot{\theta}_2^2 + \frac{m_2}{m_1+m_2} l_1 l_2 \dot{\theta}_1 \dot{\theta}_2 \cos(\theta_1 – \theta_2) \\
&\quad + l_1 g (1-\cos\theta_1) + \frac{m_2}{m_1+m_2} l_2 g (1-\cos\theta_2)
\end{aligned}
$$

It can be shown that \(V(t) \ge 0\). Its time derivative, after substituting the pendulum dynamics, simplifies remarkably to:

$$
\dot{V}(t) = -\left( l_1 \dot{\theta}_1 \cos\theta_1 + \frac{m_2}{m_1+m_2} l_2 \dot{\theta}_2 \cos\theta_2 \right) \ddot{x} – \left( l_1 \dot{\theta}_1 \sin\theta_1 + \frac{m_2}{m_1+m_2} l_2 \dot{\theta}_2 \sin\theta_2 \right) \ddot{z}
$$

To ensure swing energy dissipation, we desire \(\dot{V}(t) \le 0\). This can be achieved by designing the quadrotor drone‘s acceleration \((\ddot{x}, \ddot{z})\) to include a component proportional to the negative of the coefficients in \(\dot{V}(t)\). Therefore, we define the anti-swing component as:

$$
\boxed{
\begin{aligned}
\ddot{x}_2(t) &= k \left( l_1 \dot{\theta}_1 \cos\theta_1 + \frac{m_2}{m_1+m_2} l_2 \dot{\theta}_2 \cos\theta_2 \right) \\
\ddot{z}_2(t) &= k \left( l_1 \dot{\theta}_1 \sin\theta_1 + \frac{m_2}{m_1+m_2} l_2 \dot{\theta}_2 \sin\theta_2 \right)
\end{aligned}}
$$

where \(k > 0\) is a control gain. Substituting the total acceleration \(\ddot{x}_c = \ddot{x}_1 + \ddot{x}_2\) and \(\ddot{z}_c = \ddot{z}_1 + \ddot{z}_2\) into the expression for \(\dot{V}(t)\) yields:

$$
\begin{aligned}
\dot{V}(t) = &-\left( l_1 \dot{\theta}_1 \cos\theta_1 + \frac{m_2}{m_1+m_2} l_2 \dot{\theta}_2 \cos\theta_2 \right) \ddot{x}_1 \\
&-\left( l_1 \dot{\theta}_1 \sin\theta_1 + \frac{m_2}{m_1+m_2} l_2 \dot{\theta}_2 \sin\theta_2 \right) \ddot{z}_1 \\
&- k \left( l_1 \dot{\theta}_1 \cos\theta_1 + \frac{m_2}{m_1+m_2} l_2 \dot{\theta}_2 \cos\theta_2 \right)^2 \\
&- k \left( l_1 \dot{\theta}_1 \sin\theta_1 + \frac{m_2}{m_1+m_2} l_2 \dot{\theta}_2 \sin\theta_2 \right)^2
\end{aligned}
$$

With a sufficiently large gain \(k \ge 1/2\), the negative quadratic terms dominate, ensuring \(\dot{V}(t) \le 0\) (or at least bounded by a convergent term from the bounded \(\ddot{x}_1, \ddot{z}_1\)). This is the foundation for proving swing suppression.

Stability and Convergence Analysis

The stability analysis proceeds by applying Lyapunov and Barbalat’s lemma. From the properties of \(V(t)\) and \(\dot{V}(t)\), we can conclude that \(V(t) \in \mathcal{L}_{\infty}\), which implies \(\dot{\theta}_1, \dot{\theta}_2 \in \mathcal{L}_{\infty}\). Further analysis of the dynamics shows that \(\ddot{\theta}_1, \ddot{\theta}_2 \in \mathcal{L}_{\infty}\). Integrating \(\dot{V}(t)\) and using the boundedness of the positioning trajectory’s acceleration, we find that \(\dot{\theta}_1, \dot{\theta}_2 \in \mathcal{L}_{2} \cap \mathcal{L}_{\infty}\). By Barbalat’s lemma, this leads to the critical result:

$$
\lim_{t \to \infty} \dot{\theta}_1(t) = 0 \quad \text{and} \quad \lim_{t \to \infty} \dot{\theta}_2(t) = 0
$$

Consequently, from the definition of the anti-swing component:

$$
\lim_{t \to \infty} \ddot{x}_2(t) = 0 \quad \text{and} \quad \lim_{t \to \infty} \ddot{z}_2(t) = 0
$$

Since the positioning trajectory satisfies \(\lim_{t \to \infty} \ddot{x}_1(t) = \lim_{t \to \infty} \ddot{z}_1(t) = 0\), the total commanded acceleration also converges to zero: \(\lim_{t \to \infty} \ddot{x}_c(t) = \lim_{t \to \infty} \ddot{z}_c(t) = 0\).

The final step is to show that the swing angles themselves converge to zero. By analyzing the pendulum equations under the condition of converging accelerations and angular velocities, and applying Barbalat’s lemma again, we can prove that:

$$
\lim_{t \to \infty} \theta_1(t) = 0 \quad \text{and} \quad \lim_{t \to \infty} \theta_2(t) = 0
$$

Furthermore, we must verify that the anti-swing component does not disrupt the quadrotor drone‘s final positioning. By integrating the velocity and position equations and leveraging the results \(\lim_{t\to\infty} \theta_1, \theta_2 = 0\) and \(\lim_{t\to\infty} \int_0^t \theta_1 d\tau, \int_0^t \theta_2 d\tau = 0\) (obtained from the dynamics), it can be shown that:

$$
\lim_{t \to \infty} x_c(t) = d_x \quad \text{and} \quad \lim_{t \to \infty} z_c(t) = d_z
$$

This completes the proof that the proposed online trajectory planning law guarantees asymptotic stability for both positioning and swing suppression of the double-pendulum quadrotor drone system.

Simulation Results and Performance Evaluation

The performance of the proposed controller is evaluated through numerical simulations in the MATLAB/Simulink environment. The model parameters used for simulation are listed in the table below.

Table 1: Simulation Parameters for the Quadrotor Drone Double-Pendulum System
Parameter Symbol Value
Quadrotor Drone Mass \(M\) 10 kg
Hook Mass \(m_1\) 1 kg
Payload Mass \(m_2\) 2 kg
Upper Cable Length \(l_1\) 0.7 m
Lower Cable Length \(l_2\) 0.3 m
Anti-swing Gain \(k\) 5
Target X-Position \(d_x\) 0.6 m
Target Z-Position \(d_z\) 0.6 m
Gravity \(g\) 9.81 m/s²

The following figures illustrate the key results. The first set compares the planned trajectory for the quadrotor drone. Figure 1 shows the horizontal (x-axis) position over time. The solid line represents the final planned trajectory \(x_c(t)\) generated by the proposed controller, while the dashed line represents the pure positioning trajectory \(x_1(t)\). Crucially, both converge to the same target of 0.6 meters. The minor deviation during transit is the effect of the anti-swing component actively working to dampen oscillations. Figure 2 shows the same comparison for the vertical (z-axis) position, with both trajectories successfully reaching 0.6 meters. These plots confirm the theoretical result that the anti-swing action does not compromise the final positioning accuracy of the quadrotor drone.

The most significant results are shown in the swing angle plots. Figure 3 displays the hook swing angle \(\theta_1(t)\). Under the pure positioning trajectory (dashed line), the hook enters into persistent, nearly constant-amplitude oscillations due to the double-pendulum dynamics excited by the quadrotor drone‘s motion. In stark contrast, the proposed controller (solid line) rapidly suppresses the hook swing, bringing it to rest with minimal residual oscillation. Figure 4 shows the payload swing angle \(\theta_2(t)\). The behavior is even more pronounced here. The pure positioning command leads to large, sustained swinging of the payload. The proposed method, however, effectively dampens these oscillations, driving \(\theta_2\) to zero smoothly and quickly. This demonstrates the controller’s efficacy in handling the more challenging lower pendulum mode.

The simulation validates all theoretical claims. The quadrotor drone achieves precise positioning at \((0.6\, \text{m}, 0.6\, \text{m})\) while actively and effectively suppressing the double-pendulum swing. The anti-swing component acts as a virtual damping force, extracting energy from the pendulum system through careful manipulation of the quadrotor drone‘s acceleration profile.

Discussion and Comparative Analysis

The proposed online trajectory planning method offers distinct advantages for controlling a quadrotor drone with a double-pendulum payload. Its primary strength lies in its elegant synthesis of a baseline positioning command with a dynamically calculated swing-damping signal. Unlike pure feedback linearization or backstepping controllers, this approach does not require full-state measurement (e.g., swing angular rates) if the angles can be estimated, as the anti-swing component relies on a simple combination of angles and angular velocities. Compared to open-loop input shaping, it is more adaptive as the anti-swing component continuously reacts to the evolving state of the pendulum.

The choice of the gain \(k\) is critical. A higher gain leads to more aggressive swing damping but may result in larger transient commands for the quadrotor drone. A lower gain provides smoother drone motion but slower swing suppression. This parameter must be tuned considering the physical limits of the quadrotor drone‘s thrust and attitude control system. Furthermore, the assumption of known mass parameters \((m_1, m_2)\) is important for the anti-swing law. Robustness to parameter uncertainty is a key area for future investigation, potentially incorporating adaptive control techniques.

The method’s performance can be summarized and compared against a naive positioning-only approach in the following table:

Table 2: Performance Comparison of Control Strategies
Aspect Pure Positioning Trajectory Proposed Online Planning
Final Positioning Error Negligible (theoretically zero) Negligible (theoretically zero)
Swing Angle (\(\theta_1\)) Persistent, undamped oscillation Rapidly damped to zero
Swing Angle (\(\theta_2\)) Large, persistent oscillation Rapidly damped to zero
Control Complexity Low (open-loop) Moderate (requires swing state)
Robustness to Disturbances Low Moderate (implicit damping)
Quadrotor Drone Command Smoothness High Slightly more aggressive transients

Conclusion and Future Work

This article has addressed the complex problem of controlling a quadrotor drone tasked with transporting a payload that exhibits double-pendulum dynamics. By developing a detailed nonlinear model and proposing a novel online trajectory planning framework, we have demonstrated a control strategy that successfully decouples the objectives of precise drone positioning and active payload swing suppression. The controller’s design is rooted in a Lyapunov analysis, ensuring formal guarantees of asymptotic stability for the combined system. Simulation results confirm the theory, showing excellent performance in driving the quadrotor drone to its target while quickly eliminating oscillations in both the hook and the payload.

The work provides a solid theoretical foundation for the safe and efficient aerial transportation of suspended loads using quadrotor drone platforms. Future research directions are plentiful. Immediate extensions include implementing the controller in a three-dimensional space and integrating it with a robust attitude controller for the quadrotor drone. Investigating the use of observers to estimate swing angles and rates from drone-mounted sensors (e.g., cameras or IMUs) would eliminate the need for sensors on the cable, enhancing practicality. Furthermore, extending the method to be adaptive to unknown or varying payload masses would significantly increase its robustness for real-world applications. Finally, experimental validation on a physical quadrotor drone testbed is the essential next step to transition this promising control methodology from theory to practice.

Scroll to Top