Adaptive Control of High-Speed Aerial Vehicles Using MRAC Methods

High-speed aerial vehicles present a challenging control problem due to rapidly changing dynamics, nonlinear behaviour and environmental uncertainties. Traditional controllers rely on pre-computed parameters which lack the ability to adapt if the vehicle does not perform as desired. In contrast, adaptive control methods update parameters in real time during flight, enabling the vehicle to adjust to changing conditions and maintain performance despite limited knowledge of its dynamics. This thesis examines the performance of a high-speed aerial vehicle using one type of adaptive controller, Model Reference Adaptive Control (MRAC). The MRAC is compared to a Linear Quadratic Regulator (LQR), which uses pre-computed parameters. Two MRAC methods are investigated, a direct method based on direct updates of control parameters, and an indirect method which derives control parameters based on estimated plant dynamics. Three different uncertainty models are evaluated, based on State Space Functions, Radial Basis Functions, and Neural Networks. Concurrent Learning (CL) is also implemented to examine its performance implications. Two parameter update methods are investigated, one based on Gradient Descent (GD), and one based on a Recursive Least Squares with forgetting factor (RLS-β) algorithm. These methods are tested on two different paths, examining the performance for low and high altitudes, as well as different levels of system excitation. The results from simulations show that the direct MRAC handles the test cases well, with improved reference tracking compared to the LQR. The use of an uncertainty model further improves the performance of the system, although none clearly outperforms the others in all metrics. The direct MRAC manages to compensate for the uncertain dynamics rapidly, with minimal tracking error during aggressive manoeuvres without prior excitation. The direct MRAC is shown to also be able to adapt to different guidance parameters, whereas the LQR lacks this capability. The implemented indirect MRAC did not result in a working controller, likely due to difficulties in fitting the nonlinear vehicle to the chosen, highly simplified, reference model. The thesis concludes that a direct MRAC can provide several performance benefits to high-speed aerial vehicles. It improves performance compared to the LQR over the range of test cases, showing the capability of the direct MRAC to adapt to changing circumstances without changes to the controller. The direct MRAC method performs well despite having very little prior information about the dynamics of the plant, which proves the viability of the direct MRAC for controlling uncertain systems.