Autonomous Remote Control of Small Unmanned Aircraft

In this paper we present a novel approach to controlling small unmanned aircraft in third-person perspective scenarios. Our approach uses monocular imagery to mimic the human ability to perceive and control a remote aircraft. This allows for the autonomous control of a wide variety of aircraft with minimal on-board sensors. We improve upon other 2D single-shot detection methods by estimating a 3D box containing the aircraft, which allows us to capture a vehicle's position and attitude as measurements. Using a constant-jerk Kalman filter, we then estimate the complete aircraft state from these measurements. Finally a proportional-integral-derivative controller is used to provide closed-loop autonomous flight of an aircraft from a remote location. Hardware experiments of a completely autonomous vehicle flying waypoints illustrate the validity of this method.