Evaluation of bio-inspired techniques for flight control from an uninhabited aerial vehicle perspective

Small uninhabited aerial vehicles (SUAVs) are in high demand to cover applications that are undesirable, dangerous or even impossible to humans due to their low cost, portability and flexibility in infrastructure. These vehicles are also characterised by low wing-loading which makes them better suited for most of these applications at low altitudes and low airspeeds. However, two major challenges affecting SUAV operations are their limited on-board energy capacity and their susceptibility to turbulent and gusty conditions. These are due to limitations in scaling of electronic components, payload limitations in relation to their size, reduced cruise speed relative to wind speed and increasing angular acceleration as the size of the vehicle decreases. Innovative solutions to these challenges can be taken from the natural world. In order to reduce energy consumption during flight natural fliers make use of the environment at low altitudes with the aid of distributed sensing to react to the airflow and load distributions affecting them. The deployment and evaluation of these techniques in SUAVs constitutes the scope of this thesis. This work begins by investigating the control effort required to exploit orographic lift within urban environments. The study linked the control effort of an SUAV in simulation to gull flight observations for the same mean wind conditions. Maximum performance improvements were seen close to buildings, but came with an increased risk of collision. Distributed pressure and strain sensors around the wing of an SUAV were investigated to help reduce this risk and to further improve performance. An initial study into the application of these sensors for end-to-end airspeed control by means of artificial neural networks (ANNs) was described. Results suggested that the combination of novel and conventional sensors produced the best results overall, even when tested with different Dryden disturbance levels not included in the training set. This study served as a proof of concept for the use of ANNs for end-to-end flight control using pressure and strain sensor information. The distributed sensing array was then used to estimate and control the aerodynamic coefficients in simulation and in wind tunnel experiments. A series of reinforcement learning agents were trained to track and maximise lift coefficient, and to maximise lift over drag ratio. These techniques were shown to be an effective means of directly controlling aerodynamic parameters and to improving manoeuvrability and flight efficiency. If applied to outdoor flight of an SUAV, the findings of this work could potentially increase flight endurance, range and efficiency, as well as improve the manoeuvrability of the vehicle. An SUAV could utilise the techniques described in this thesis to extend its missions, to increase its safe flight envelope and to rapidly avoid collision.