Introduction
Quadrotor unmanned aerial vehicles (UAVs) have become indispensable in various applications, from surveillance to delivery. However, achieving precise trajectory tracking remains a significant challenge due to the system's inherent nonlinearities, underactuated dynamics, and susceptibility to external disturbances. Traditional control methods often struggle with physical constraints like thrust saturation and mathematical issues such as singularities in Euler angle representations.
Problem Formulation
Controlling a quadrotor involves navigating three primary obstacles that often bridge the gap between theoretical stability and real-world implementation:
- Mathematical Singularities: Most controllers utilize Euler angles (roll, pitch, yaw). When the pitch or roll angles approach $\pm\pi/2$, the mathematical model encounters a singularity, commonly known as "gimbal lock," leading to system failure.
- Thrust Saturation: Real-world motors have finite physical limits. If a control law requests thrust beyond these limits, the system can become unstable. Many theoretical models overlook these saturation effects.
- Computational Complexity: Standard "backstepping" techniques often require calculating high-order derivatives of the desired attitude, which is computationally prohibitive for real-time applications.
Proposed Control Strategy
To address these issues, this research proposes a two-loop (position and attitude) adaptive control structure using several specialized techniques.
Differentiable Saturated Function
To manage thrust limits, a new smooth, differentiable saturation function is introduced. Unlike standard "hard-clipping" methods, this differentiable approach allows for rigorous stability analysis using Lyapunov theory, ensuring the translational thrust remains within a preassigned safe range in the position control loop.
Singularity-Free Torque Control
The proposed attitude control loop avoids complex analytic derivatives of the desired attitude. Instead, it utilizes a compensation term based on an attitude error function. This simplification ensures the quadrotor remains within a safe operational zone ($|\phi|, |\theta| < \pi/4$) and eliminates the risk of gimbal lock without the need for computationally heavy "command filters."
Adaptive Parameter Estimation and BLFs
Since physical parameters like mass ($m$), inertia ($I$), and rotor constants ($J_r$) are often uncertain or change over time, the controller incorporates online estimators. Furthermore, Barrier Lyapunov Functions (BLFs) are employed to guarantee that the system states never violate predefined boundaries; the BLF value approaches infinity as the system approaches a constraint, effectively preventing the quadrotor from hitting mathematical singularities.
Key Insights: The Technical Lemma
A pivotal contribution of this work is a mathematical lemma that establishes a formal relationship between attitude tracking error and position tracking error under thrust saturation. This theoretical foundation ensures that the position and attitude control loops operate harmoniously, even when the motors are pushed to their physical limits.
Simulation Results and Performance
The effectiveness of the strategy was validated through simulations based on a real-world quadrotor model. Key findings include:
- Asymptotic Convergence: Tracking errors for both position ($x, y, z$) and attitude ($\phi, \theta, \psi$) converged exactly to zero, demonstrating superior performance over standard Proportional-Derivative (PD) controllers.
- Constraint Satisfaction: Roll and pitch angles were strictly maintained within the safe $(-\pi/2, \pi/2)$ range throughout the flight, even under noise.
- Thrust Management: Translational thrust was successfully capped (e.g., at $22.0672$ N), preventing motor saturation-induced instability.
- Robustness: The system remained stable and accurate even in the presence of time-varying uncertain noise in both position and attitude dynamics.
Conclusion and Future Work
This paper presents a robust framework for quadrotor control that respects both physical and mathematical constraints. By combining adaptive parameter estimation with a singularity-free torque controller, the authors achieve asymptotic tracking without the computational burden of traditional backstepping. Future research will focus on physical hardware implementation to account for sensor latency and exploring finite-time convergence to force errors to zero within a predefined timeframe.
Summary updated