Robust and Intelligent Control of Unmanned Aerial Vehicles

Abstract

Quadrotors from the family of unmanned aerial vehicles have an important place for human life, because over the last decade, they have been used in many areas both civilian and military applications, such as search, surveillance, rescue, tracing, aerial photography and postal service due to their size and maneuverability. Therefore, there are a great amount of the studies about the modelling and control of the quadrotors in the literature. Despite all these efforts, the modelling and control of the quadrotors is still among the subjects which are frequently studied to make them more autonomous. What makes them so important is that they have hover, vertical take-off and landing VTOL ability and agile mobility. With these features, even complex tasks can be successfully accomplished.

Quadrotor that is an under-actuated and nonlinear coupled system, has four rotors and six degrees of freedom (6 DOF) involving the both translational and rotational dynamical equations. Its unstable nature has required many different control methods. The most remarkable control methods among them are optimal control, robust control, adaptive control and intelligent control. The main goal of these control strategies is to achieve the best performance in the quadrotor control. However, there are many factors that affect the performance of the quadrotors such as unmodelled dynamics, parameter uncertainties, all external force and moment disturbances, payload changes and sensor measurement noises during the quadrotor flights. In order to deal with these factors, many linear and non-linear controllers including above control strategies have been developed. Disturbance/Uncertainty Estimator (D/UE) based control, or in other words, disturbance observer based control (DOBC) that compensates the external disturbances and system uncertainties is one of the efficient robust control approaches and they are frequently used in modern control systems. In this thesis, widely used DOBC approaches in the literature are discussed in detail, usage structures for quadrotor control architectures are studied and a new machine learning assisted DOBC approach is proposed.

This thesis study can be summarized in three main subjects. Firstly, an analysis and synthesis of widely used linear disturbance observer based robust control approaches are presented. The main objective is to provide an exhaustive comparison of disturbance observer based robust control approaches and to handle the structural details of each approach for gaining insight about the complexity of each approach. Toward this goal, nine performance and robustness equations portraying useful insights for understanding and analyzing control systems are derived by examining their common and equivalent block diagrams. Four of them have been selected as a Gang of Four (GoF) equations, namely, Complementary Sensitivity Function (CSF), Sensitivity Function (SF), Disturbance Sensitivity Function (DSF) and Noise Sensitivity Function (NSF). Robustness and disturbance rejection performance analysis of all linear disturbance observer based control schemes and Classical Feedback Control (CFC) scheme are done using GoF equations. With these representations, two tables discussing all prime issues and facilitating the selection of the best approach have been obtained. Our research stipulates critical facts and figures of each scheme by considering the derived GoF equations, which can be used for choosing the most appropriate disturbance observer based control approach for a given robust control problem. It is concluded that the Uncertainty Disturbance Estimator (UDE) approach is superior when time delay type uncertainty is involved in the model. Unfolding this is critical as time delay is an inevitable fact in most industrial control systems. The findings also emphasize that Time Domain Disturbance Observer Based Control (TDDOBC) approach is proficient if there is no process time delay.

Secondly, we present a short tutorial introduction to disturbance observer based control approaches for the quadrotors. With this tutorial, researchers, engineers and students would be able to implement disturbance observer based model-in-loop simulations and experiments more easily to design robust autopilot system for the quadrotors. To achieve this, the modeling and controlling of a quadrotor are explained and all linear disturbance observer based control approaches in the literature are adapted its overall nonlinear architecture. Disturbance observer based control design steps are given in detail by design challenges. To show their disturbance rejection capabilities and practical applicability, two flight simulation scenarios are carried out. For all simulation cases, we only take into account the external disturbances in rotational motions. While we give the attitude trajectory commands to quadrotor attitude control architecture in the first scenario, we issue both way-point and trajectory commands to an outer loop controlling the translational motions in the second one. Presented disturbance observer based control approaches have successfully completed the given reference commands in the presence of the external disturbances even under the measurement noise. Moreover, simulation experiments have shown that UDE approach transmit the external disturbance and measurement noise effects to the actuators directly. As a result, for UDE approach, it should be kept in mind that flight accidents may occur due to excessive electronic speed controller heating. Baseline attitude controller without disturbance observer based control approach have failed to follow the given reference commands. The simulation studies have also proved the practical applicability of these methods, which are successful even under measurement noise.

As the final and main purpose, we introduce a machine learning assisted disturbance/uncertainty estimator based control scheme. The aim of the proposed method is to update the nominal model directly used by the conventional disturbance observer based control architecture and approximate it to the perturbed/uncertain system using machine learning approaches. This enhances the disturbance rejection performance of the system remarkably. The performance deterioration capacity of lumped disturbances, which are the mixed effect of disturbances entering through the control channels and modeling uncertainties, are decomposed in our approach and handled separately. For this study, harmonic disturbance model and constant unstructured uncertainty model are considered, and ϵ-Support Vector Regression approach is used together with an online adaptation algorithm. A numerical example is given to demonstrate the merits and effectiveness of the proposed approach. Simulation results show that the proposed method outperforms the conventional disturbance/uncertainty estimator based control architecture by increasing disturbance estimation performance of the system.