Autonomous soaring flight for unmanned aerial vehicles

Abstract

Unmanned Aerial Vehicles (UAVs) provide unique capabilities in a range of industrial, scientific and defence applications. A small UAV could extend flight duration without requiring additional propulsive power through the use of soaring. This thesis examines the aerodynamic mechanisms of soaring flight and proposes planning and control algorithms for a UAV to autonomously sense and utilise the wind environment to extend flight duration. In order to utilise soaring a thorough understanding of the energy interaction between an aircraft and the surrounding atmosphere is required. This thesis presents a mathematical model for a gliding aircraft and examines how wind contributes to the energy change of an aircraft. Conditions for optimal energy efficiency are identified for gliding and soaring flight in linear wind shear. The proposed path planner takes advantage of the energy equations for a gliding aircraft to plan energy efficient paths over a known wind field. Previous soaring planners have focused on a single type of energy gain such as static soaring. By using the energy equations directly the planner can exploit all energy gain conditions rather than relying on specialised controllers. The planner requires an adequate estimate of the wind field to plan reliable energy gain paths. A small UAV would typically only have access to direct wind observations taken during flight. Gaussian Process (GP) regression is proposed to generate a wind map from direct wind observations. This model-free approach can account for static and dynamic wind fields and does not restrict the planner to particular types of wind structure. Maintaining an accurate map requires the planner to ensure efficient map sampling and maintain sufficient energy to continue flight. The path planning algorithm exploits the variance estimate from the GP map to identify regions of the map which require improvement. The planner assesses the aircraft’s energy state and current map to determine target regions of the wind field for further exploration or energy exploitation. Results demonstrate that this architecture is capable of generating energy-gain paths in both static and dynamic wind fields. The mapping algorithm records direct samples of the wind to generate a wind map that is used by the planning algorithm to simultaneously explore and exploit the wind field to extend flight duration without propulsive power.