UAV Parameter Estimation with Gaussian Process Approximations

Abstract

Unmanned Aerial Vehicles (UAVs) provide an alternative to manned aircraft for risk associated missions and applications where sizing constraints require miniaturized flying platforms. UAVs are currently utilised in an array of applications ranging from civilian research to military battlegrounds. A part of the development process for UAVs includes constructing a flight model. This model can be used for modern flight controller design and to develop high fidelity flight simulators. Furthermore, it also has a role in analysing stability, control and handling qualities of the platform. Developing such a model involves estimating stability and control parameters from flight data. These map the platform's control inputs to its dynamic response. The modeling process is labor intensive and requires coarse approximations. Similarly, models constructed through flight tests are only applicable to a narrow flight envelope and classical system identification approaches require prior knowledge of the model structure, which, in some instances may only be partially known. This thesis attempts to find a solution to these problems by introducing a new system identification method based on dependent Gaussian processes. The new method would allow for high fidelity non-linear flight dynamic models to be constructed through experimental data. The work is divided into two main components. The first part entails the development of an algorithm that captures cross coupling between input parameters, and learns the system stability and control derivatives. The algorithm also captures any dependencies embodied in the outputs. The second part focuses on reducing the heavy computational cost, which is a deterrent to learning the model from large test flight data sets. In addition, it explores the capabilities of the model to capture any non-stationary behavior in the aerodynamic coefficients. A modeling technique was developed that uses an additive sparse model to combine global and local Gaussian processes to learn a multi-output system. Having a combined approximation makes the model suitable for all regions of the flight envelope. In an attempt to capture the global properties, a new sampling method is introduced to gather information about the output correlations. Local properties were captured using a non-stationary covariance function with KD-trees for neighbourhood selection. This makes the model scalable to learn from high dimensional large-scale data sets. The thesis provides both theoretical underpinnings and practical applications of this approach. The theory was tested in simulation on a highly coupled oblique wing aircraft and was demonstrated on a delta-wing UAV platform using real flight data. The results were compared against an alternative parametric model and demonstrated robustness, improved identification of coupling between flight modes, sound ability to provide uncertainty estimates, and potential to be applied to a broader flight envelope.