Low-Altitude Control and Local Re-Planning Strategies for Small Fixed-Wing UAVs

In recent years, small, easily manageable, operated, and maintained Unmanned Aerial Vehicles (UAVs) have become ubiquitous in an every-growing set of industrial, humanitarian, scientific, and commercial domains. For large-scale remote sensing and mapping applications, small fixed-wing platforms provide the advantages of longer range and higher speeds, with respect to their Micro-Aerial Vehicle (MAV) counterparts. However, today’s fixed-wing UAVs are largely limited to a primitive set of basic waypoint following and pre-programmed tasks, with little awareness of the environment in which they fly, adaptability to changing conditions, or higher-level decision making capabilities. Particularly relevant to small fixed-wing mission profiles, is the required ability to operate safely near uncertain terrain while disturbed by possibly strong and turbulent wind fields. Enabling these activities entails the design of efficient, robust, and more adaptable motion planning and control algorithms which moreover adhere to the vehicle’s restrictive dynamic flight envelope. The primary goal of this thesis is to develop practical control and local re-planning strategies for low-flying, small, fixed-wing UAVs with explicit awareness of these environmental hazards. Part A of this thesis addresses the challenge of guiding small, low-speed fixed-wing aircraft in strong winds. Our first contribution in this part is the unique consideration of excess winds, i.e. wind speeds that exceed the vehicle’s nominal airspeed, within a lateral-directional control law for fixed-wing UAVs. We develop a principled, nonlinear guidance law which guarantees convergence to a safe and stable vehicle configuration with respect to the wind field while preserving some tracking performance with respect to the path target. We then expand on this concept by including an energy efficient airspeed reference compensation logic, enabling not only mitigation, but also prevention or over-powering of excess winds which would otherwise cause the aircraft to "run-away". We emphasize heavily in this second iteration on field testing results, demonstrating track keeping errors of less than 1 meter consistently maintained during gusting excess winds over various mountainous regions in Switzerland. The third component of this part revisits the efficiency of the airspeed references. A coupled approach to airspeed and heading reference commands is developed with a more principled consideration of airspeed reference minimization. The coupled method is compared against the previous decoupled approach in simulations showing both increased power-efficiency and tracking performance in static and dynamic winds. Part B of this thesis operates as a road map to fast, environment-aware local re-planning for fixed-wing UAVs operating near terrain. We first delve into the practicalities of deploying a guidance level Nonlinear Model Predictive Controller (NMPC) on a small fixed-wing platform. We develop new control augment modeling methods featuring reduced order models of the underlying low-level autopilot response and quasi-steady forces, simple parameter identification procedures, and open loop predictability on the order of tens of seconds, making our modeling approach suitable for long horizon NMPC. Through flight experiments, we demonstrate Dubins aircraft path segment tracking in three dimensions with wind speeds exceeding 50% of the vehicle’s airspeed, and further show a mock motor failure scenario. A particular focus is further spent on soft constraint formulation. The third component of this part reworks the developments of the first two towards our local re-planning formulation. Wind-aware reference trajectory generation is developed from the guidance logic in Part A for lateral-directional states, and vertical wind is included in the longitudinal guidance. Vision-based elevation mapping is utilized to to provide a generalized 2.5D world representation to the aircraft. The map is bilinearly interpolated for height feedback, and we design an efficient ray casting approach for detection of forward (line of flight) and lateral occlusions. The occlusions are used to construct novel "relative" Euclidean Signed Distance Fields (RESDFs), which are a function of the relative velocity between the vehicle and obstacle. We further present a method of transforming the RESDFs into optimizable soft constraints for the objective function of the NMPC. A preliminary example of the full system acting to avoid an obstructing hillside is demonstrated in hardware-in-the-loop (HITL) simulation