Design and experimental validation of an adaptive integral chattering-free sliding mode control for quadrotor attitude tracking

Abstract

Quadrotor attitude control must remain accurate and stable despite modeling uncertainties, external disturbances, and partial actuator faults—conditions that routinely degrade performance in practice. Classical sliding mode controllers are robust but suffer from chattering, while PID controllers are simple but lack robustness to bias and uncertainty. This thesis aims to achieve accurate, smooth, and robust attitude regulation by designing and validating a controller that preserves robustness without chattering and maintains performance under uncertainty and faults. A physics-grounded yet control-oriented model is developed by defining inertial and body-fixed frames, specifying attitude with a standard yaw–pitch–roll sequence, and deriving rigid-body kinematics and Newton–Euler dynamics that capture thrust/torque generation, translational drag, and gyroscopic couplings. For controller design, well-justified simplifications are introduced—small attitude-channel aerodynamic moments and rotor inertia are neglected, while dominant gyroscopic effects are retained in compact form—to isolate the attitude subsystem without losing key behaviors. The result is a compact, reproducible attitude model that is affine in the control inputs and well suited to robust and adaptive synthesis. Four controllers are evaluated side by side: a tuned PID, a conventional sliding mode controller (SMC), a chattering-free full-order SMC (FOSMC), and the proposed Adaptive Integral Chattering-Free SMC (AICFSMC). PID is attractive for its simplicity and smooth actuation, but it lacks inherent robustness—persistent biases, parametric mismatch, and coupled disturbances readily produce steady-state errors and degraded tracking. Conventional SMC offers strong robustness, yet its discontinuous switching generates chattering that excites unmodeled dynamics, increases control RMS, and can harm actuators. Literature FOSMC variants mitigate chattering via continuous control laws and higher-order design, but they do not include online adaptation; as a result, they remain sensitive to unknown, time-varying uncertainties or biases and cannot systematically compensate when the plant deviates from its nominal model. This thesis introduces an Adaptive Integral Chattering Free Sliding Mode Controller that achieves chattering suppression by computing the first order time derivative of the control signal, integrating it, and applying the resulting continuous input to the plant, thereby removing discontinuous switching while preserving robustness; on top of this, an online parameter adaptation estimates only the required components of the affine lumped dynamics so that modeling errors and disturbances are compensated as needed; a Lyapunov-based analysis guarantees bounded internal signals and asymptotic convergence of the tracking errors, with the full construction developed on the roll channel and then applied uniformly to pitch and yaw for a consistent three axis design. Controllers are implemented in MATLAB/Simulink with a fixed-step fourth-order Runge–Kutta solver and a 10 ms sample time. A single quadrotor attitude model and identical reference commands are used across all runs to ensure fair comparison, with controller and vehicle parameters kept constant throughout. Three scenarios are evaluated: (i) nominal operation without disturbances or faults, (ii) robustness with time-varying, non-zero-mean disturbance torques and a 30\% uncertainty applied to the inertia terms, and (iii) fault tolerance with a 10\% loss of effectiveness injected into one actuator at a known time. Performance is assessed using time-domain tracking behavior, integral error measures, and control-effort statistics such as RMS, with signals logged over identical horizons and repeated trials to confirm repeatability. In the nominal scenario, all controllers track the attitude references closely; the primary difference is actuation quality. Conventional SMC produces visibly oscillatory inputs with the largest control RMS due to switching, whereas PID, FOSMC, and AICFSMC generate smooth signals of much lower magnitude; AICFSMC matches the best smoothness while preserving precise tracking. Under disturbances and 30\% inertia uncertainty, PID and conventional SMC develop noticeable steady-state offsets and degraded transients. FOSMC mitigates oscillations and remains smooth but still shows measurable deviations when disturbances persist. AICFSMC maintains accurate tracking across roll, pitch, and yaw while keeping control effort moderate and free of high-frequency content, indicating effective online compensation of the lumped uncertainty without discontinuous switching. With a 10\% actuator loss of effectiveness, except for proposed controller all the controllers exhibit sustained errors and larger deviations after the fault. AICFSMC adapts to the reduced authority and preserves accurate tracking without post-fault drift, with continuous, bounded inputs. Across scenarios, AICFSMC consistently delivers the best combination of accuracy, robustness to uncertainty and faults, and actuator-friendly control effort. All controllers are auto-generated from Simulink to C/C++ and integrated as PX4 modules running on a real-time operating system with priority-based scheduling; messaging uses a publish/subscribe bus for deterministic sensor–actuator timing. I/O is first verified with Connected I/O, then External Mode executes on-target while streaming signals for live monitoring and gain tuning; a standard mixer maps normalized attitude commands to per-motor outputs under enforced limits. The hardware platform is a DJI F450 with a matched propulsion set and a Cube Orange+ flight controller; experiments use a pivoting stand that frees the roll axis while mechanically locking pitch and yaw. In this setup, the proposed AICFSMC tracks sinusoidal roll references with minimal lag and the lowest tracking error, delivering continuous inputs without chattering. Conventional SMC shows irregular actuation and occasional divergence despite tuning, and PID exhibits noticeable phase delay. Control-effort statistics confirm that AICFSMC maintains smooth, bounded actuation comparable to the best alternatives while avoiding high-frequency content. Adaptation signals remain well-behaved and effectively compensate unmodeled rig effects, indicating stable online parameter updates and robust closed-loop behavior in real time. This thesis makes two primary contributions and demonstrates their practical significance. First, it proposes an Adaptive Integral Chattering-Free Sliding Mode Controller that eliminates chattering by computing the first-order time derivative of the control signal to be integrated before application to the system, while employing online parameter adaptation to estimate only the necessary components of the affine lumped uncertainty. Second, it establishes the controller's effectiveness through an end-to-end validation pipeline, ranging from reproducible simulations under nominal, uncertain, and faulty conditions to embedded experiments on a standard autopilot with model-based code generation. Collectively, the results demonstrate accurate, smooth, and robust attitude regulation in the presence of uncertainties, disturbances, and actuator faults, with straightforward integration into existing flight stacks. These properties point to a clear path for extending the method to full six-degree-of-freedom flight with outer-loop position control and motivate future directions such as finite-time formulations, automated gain tuning, online fault diagnosis, and outdoor trials in realistic wind and mission scenarios.