Nowadays, new designs and configurations of small-scale unmanned aerial vehicles (UAVs) are being developed to meet the higher requirements of power efficiency and maneuverability for some professional fields, such as search and rescue, parcel delivery, line inspection, surveying, surveillance and reconnaissance (ISR), etc. To achieve these requirements, a simple way is to combine a quad-rotor and fixed-wing to generate a new type of aircraft called vertical take-off and landing (VTOL) UAVs. There are many research results published on design, aerodynamic model, dynamic model, and control of VTOL UAVs, but no one has successfully developed a mature and reliable VTOL UAV system. Therefore, our group proposed a project aimed to design, analyze and build a reliable VTOL UAV system,...[ Read more ]

Nowadays, new designs and configurations of small-scale unmanned aerial vehicles (UAVs) are being developed to meet the higher requirements of power efficiency and maneuverability for some professional fields, such as search and rescue, parcel delivery, line inspection, surveying, surveillance and reconnaissance (ISR), etc. To achieve these requirements, a simple way is to combine a quad-rotor and fixed-wing to generate a new type of aircraft called vertical take-off and landing (VTOL) UAVs. There are many research results published on design, aerodynamic model, dynamic model, and control of VTOL UAVs, but no one has successfully developed a mature and reliable VTOL UAV system. Therefore, our group proposed a project aimed to design, analyze and build a reliable VTOL UAV system, which can make emergency deliveries of blood samples, first-aid medicine, etc. This project was finished successfully with our group's effort by 2018, and a comprehensive solution for the VTOL UAV systems was provided.

In this thesis, we will first present an overview of the whole project, then focus on the main contributions of this thesis, which are theoretical analysis and practical implementation of a unified control framework for controlling a quadrotor tail-sitter UAV. The most salient feature of this framework is its capability of uniformly treating the hovering and forward flight, and enabling continuous transition between these two modes, depending on the commanded velocity. The key part of this framework is a nonlinear solver based on a sequential convex programming (SCP) algorithm, called SCP solver, which solves for the proper attitude and thrust that produces the required acceleration set by the position controller in an online fashion. Then the planned attitude and thrust are achieved by an inner attitude controller. Besides planning the attitude and thrust required by the position controller in an online fashion, this framework could also be used off-line to analyze the UAV's equilibrium state (trimmed condition), especially when wind gust is present.

Firstly, aircraft design and system modeling of the tail-sitter VTOL UAV is presented, and the system modeling includes aircraft dynamics, kinematics, actuator dynamics, and aerodynamics. Then, a unified control framework based on the SCP solver is introduced in details, and the effectiveness is demonstrated by an ideal plant at the first stage. To further verify this control method, a realistic simulator is developed based on the system modeling, and extensive flight trajectories including hovering, takeoff, transition, forward flight, landing, circling and even high maneuverable trajectories are designed to test the performance of this method. The testing results are satisfactory with a small tracking error and smooth attitude angle change for all flight modes, but this method can not be implemented on the real platform directly because the computation time of the SCP solver is too long.

To address this long computation time problem of the nonlinear solver, a novel idea is inspired by the widely used imitation learning techniques in recent years. This idea is to apply imitation learning techniques to learn the behavior of the SCP solver, and then generate a neural network called SCPNet to replace the SCP solver for real-time implementation. Besides the low-computation cost, the proposed learning method, with proper dynamics excitation and data collection during the training, can approximate the nonlinear solver solutions accurately.

As the tail-sitter VTOL UAV is susceptible to significant model uncertainty, various flexible modes, and complicated aerodynamic damping effects, a robust controller for the control structure's inner loop (i.e., angular rate loop) is designed and implemented based on model identification techniques. Finally, extensive indoor and outdoor experiments are presented to demonstrate the performance of this unified control framework and illustrate the stability, robustness, and superiority of this tail-sitter VTOL UAV system.