Model predictive based UAV formation flight control : formulation, extension and experiment.
Date of Issue2012
School of Mechanical and Aerospace Engineering
This thesis considers the problem of unmanned aerial vehicle (UAV) formation flight control. This problem has received significant attention recently in the control, robotics and UAV community due to its numerous potential applications ranging from simple to complex tasks. The focus of this thesis is on the development of formation flight control in hierarchical fashion based on model predictive control (MPC) approach. This approach is interesting in its own right and in this thesis three novel variants of MPC formation flight control approaches are developed. Each variant is developed to achieve improvement in terms of performance, robustness and/or safety. The first variant utilizes multiplexed MPC to convert the computationally-heavy centralized formation flight control problem into computationally-feasible decentralized ones while addressing the robustness issue. Under this method, the whole centralized formation flight problem is divided into decentralized subsystems. One major advantage of this method is that the closed-loop stability of the whole formation flight system is always guaranteed even if a different updating sequence is used, which makes the scheme flexible and able to exploit the capability of each UAV fully. The obstacle avoidance scheme in MPC formation flight control is extended from the literature. By combining the spatial horizon and the time horizon, the small pop-up obstacles avoidance is transformed into additional convex position constraint in the MPC online optimization. Considering the complexity of the UAV dynamics and in order to realize the formation flight, the second formation flight control variant proposes the combination of MPC and robust feedback linearization. A specific type of UAV, i.e. the quadcopter is considered in the formation flight. In order to achieve real-time MPC optimization, two modifications, i.e. the control input hold and variable prediction horizon, are made. Formation flight experiments are set up in the Vicon environment and the flight results demonstrate the effectiveness of the proposed formation flight architecture. The last formation flight control variant proposes the combination of MPC with adaptive neural networks. In order to make the formation flight safer during extreme maneuvers, the reachability algorithm based on the safe maneuver envelope is proposed in the MPC framework. The “Pseudo-Control Hedging” technique is also implemented in the MPC formation flight planner resulting in the tight coupling between the attitude adaptive control and the MPC planner. Formation flight simulations are then conducted based on detail-modeled quadcopter in Simulink/Matlab. In this work, the formation flight algorithms are validated throughout a series of simulations and for the cases where hardware realization is feasible, experiments using quadcopters as the test-bed vehicles. Technical challenges as well as recommendations for future work are highlighted.