Teaching micro air vehicles how to fly as we teach babies how to walk

Abstract

Recently, various micro air vehicles have drawn significant attention in numerous areas including surveillance and reconnaissance. The manual control of micro air vehicles is very difficult due to their smaller profile; therefore, a stability and controllability augmentation system is a minimum requirement for stable and efficient flight. However, it is not easy to obtain an accurate numerical model for the flight dynamics of micro air vehicles in the design of the stability and controllability augmentation system. An alternative approach for the stability and controllability augmentation systems is to incorporate reinforcement learning in order to address the numerical complexity. However, in order to train micro air vehicles to learn how to fly, they must first be airborne. This article presents a new method that provides an effective environment where a micro air vehicle can learn to fly in a similar manner to an infant learning to walk. The test setup was constructed to enable the magnetic levitation of a micro air vehicle that has a permanently embedded magnet. This apparatus allows for flexible experimentation: the position and attitude of the micro air vehicle, the constraint forces, and the resulting moments are adjustable and fixable. This "Pseudo Flight Environment" was demonstrated using a fixed-wing micro air vehicle model. Furthermore, in order for the model to maintain a constant altitude, a height hold control system was devised and implemented.