"Unmanned aerial vehicles (UAVs) have been increasingly proposed for aerial surveillance, mapping, and delivery tasks. Historically these vehicles fall into two categories: conventional fixed-wing aircraft, which are capable of efficient flight over long distances but lack maneuverability, and rotorcraft, which are capable of agile and maneuverable flight but lack efficiency and endurance. Recent advancements in aerial vehicle design aim to incorporate characteristics from both rotorcraft and conventional fixed-wing aircraft, ultimately creating aircraft that are capable of both maneuverable and efficient long distance flight. These type of platforms are ideal for tasks that require both the ability to maneuver through cluttered environments, and the ability to fly long distances efficiently. An aircraft of this type, the agile fixed-wing aircraft, is a fixed-wing aircraft characterized by a high thrust-to-weight ratio (> 1), and large control surfaces capable of large deflections.The objective of this thesis is to further the autonomous capabilities of agile fixed-wing aircraft; specifically in the context of control systems and real-time collision avoidance. The thesis begins with a discussion of a previously developed flight dynamics model, and presents a method for validating a flight dynamics model in flight regimes that rely on feedback control. Subsequently, a single control architecture is developed that can track trajectories within both conventional and aerobatic flight regimes. This architecture is then extended to be applicable to many other types of vehicles, specifically vehicles which can generate a torque in an arbitrary direction, and can apply a single body-fixed force. We demonstrate autonomous aerobatic trajectories with an agile fixed-wing aircraft, specifically knife-edge, rolling harrier, aggressive turnaround and hovering maneuvers within conventional simulations, hardware-in-the-loop simulations, indoor flight tests and outdoor flight tests. We also validate the extension to other platforms by demonstrating flips with a quadrotor in both simulation and outdoor flight tests. All flights were performed with on-board sensing and computation.We then present a reactive obstacle avoidance algorithm that utilizes the maneuvering capabilities of agile fixed-wing aircraft and can be run in real-time with on-board sensing and computation. At each time step, trajectories are selected in real-time from a pre-computed library that lead to various positions on the edge of the obstacle sensor's field-of-view. A cost is assigned to each collision-free trajectory based on its heading toward the goal and minimum distance to obstacles, and the lowest cost trajectory is tracked. If all of the potential trajectories leading to the various positions at the edge of the obstacle sensor's field-of-view result in a collision, the aircraft has enough space to hover and come to a stop, which theoretically guarantees collision-free flight in unknown static environments. Autonomous flight in unknown and unstructured environments using only on-board sensing (stereo camera, IMU, and GPS) and computation is demonstrated with an agile fixed-wing aircraft in both simulation and outdoor flight tests. During the flight testing campaign, the aircraft autonomously flew 4.4 km in a tree-filled environment with an average speed of 8.1 m/s and a top speed of 14.4 m/s"--